Calibrating 3458A?

[Dr. Frank]

no, the NVRAM data do not change over time.

That is not quite correct...

The calibration constants only change when calibration procedures are followed, but the NVRAM is used to store a couple of undocumented variables which change outside calibration context.

The `CAL?` command has at least the following undocumented uses:

        # Undocumented CAL? queries
        # CAL? -1       21249=NUM_SAMPLES
        # CAL? -2       861=BURST2
        # CAL? -3       1149=BURST1
        # CAL? -4       11=NREADINGS
        # CAL? -5       2010=SMPL_RATE
        # CAL? -6       1=SSMP_INC
        # CAL? -7       3.54147092E-05=PERIOD
        # CAL 2437      16611, DEFEATS valid 2437
        # CAL 2941      0, DESTRUCTIVE OVERLOADS valid 2941

The last two of which are stored in the NOVRAM

Hello Poul-Henning,
thank you again for uncovering these secret parameters.

I simply took the question literally "do the data change over time?",  , i.e. an nvRAM is intended to keep its content stable over at least 10 years.
The data of course change over a write operation, but that's a different story. 

Btw.: We still have not tested the ADR1000 inside the 3458A for lower noise prformance, yet.
You kindly assisted when I changed the CAL limit constants in the EPROM.

Frank

[Dr. Frank]

Dr Frank, here are some photos. Looks like the Vishay 40k resistor is there (R207). The IC datecodes are nearly all '95.

Keith

Hello Keith,
you got FW 8, so that probably was originally installed in 1995 already, at least the EPROM and the solder joints look original. That's fully ok, no need to update to V 9, as all of the severe bugs are fixed with V8 already.

The type designator of R207 should be on the back side, but probably there's only 40k000 and 0.1% imprinted, then you'd still have the older resistor version inside. I've got the VHP101, which is rock stable over time (a decade), and has quite a low T.C. , as far as I can tell.

Frank

[bsdphk]

thank you again for uncovering these secret parameters.

What can I say, the HP3458 firmware is some of the most interesting I have disassembled yet :-)

(Disassembling old code is a computer-archaeology hobby of mine, and my HP instruments have made for good research material: https://github.com/bsdphk/PyReveng3)

Since the HP3458 is a current product, I can obviously not publish the (tools for) reverse-engineering that.

[Keith956]

you got FW 8, so that probably was originally installed in 1995 already, at least the EPROM and the solder joints look original. That's fully ok, no need to update to V 9, as all of the severe bugs are fixed with V8 already.

I had thought from a cursory look that it was Rev 9 but you're right, that 9 is really an 8.

I'll take a closer look at R207 next time I've got the top off, which will be when the cal ram needs to be replaced.

[Dr. Frank]

you got FW 8, so that probably was originally installed in 1995 already, at least the EPROM and the solder joints look original. That's fully ok, no need to update to V 9, as all of the severe bugs are fixed with V8 already.

I had thought from a cursory look that it was Rev 9 but you're right, that 9 is really an 8.

I'll take a closer look at R207 next time I've got the top off, which will be when the cal ram needs to be replaced.

Sorry, I have to correct myself. I found the part number of your FW version @ xdevs. 03458-88887 is FW 6 only.
That makes much more sense concerning its vintage.

So I recommend to upgrade to V 8 or 9, when you change the nvRAMs.
Single EPROM binary is available, maybe not on xdevs 
Either buy a new EPROM, 27C4002-100, I see it for expensive 11€ at Reichelt, Germany.
Or erase the old one if you got the equipment.

Frank
 

[Keith956]

Is that correct?

If I use the REV? command from the front panel it returns : 8,2

And the EEPROM has Rev=8 on it.

[Dr. Frank]

Yes, you're correct, of course. And I just got my own original replaced EPROM, V8.

It's wrong here: https://xdevs.com/fix/hp3458a/#firmware.
If you scroll down a little bit there, TiN has put some information about the version of that R207.



 

[martinr33]

Here's a version of the V9 software that programs directly into a 16-bit EPROM. The original version I downloaded was byteswapped.

The checksum of this version is 03FCB1D6.

https://www.dropbox.com/s/7h7pqodpv5ifftm/output_rev9.bin%20-%20CORRECTED%2003FCB1D6.bin?dl=0

 

[alm]

Looks like the Vishay 40k resistor is there (R207).

Where do you see the VHP101 markings?


I believe this is a picture of the VHP101 resistor (by TiN from here):


And this is of the old resistor they used before they switched to the VHP101 (by TiN from here):

[TiN]

301031 is still VHP resistor, just custom order P/N. I have not seen evidence that this older marking resistor is any worse than standard VHP101 found in newer units.

We still have not tested the ADR1000 inside the 3458A for lower noise prformance, yet.

Bit info on this. I can say that noise performance so far unchanged with ADR instead of original A9 in my unit. This lead me to think more into A3 modifications, but due to other ppm activities didn't resume those experiments yet.

[alm]

301031 is still VHP resistor, just custom order P/N. I have not seen evidence that this older marking resistor is any worse than standard VHP101 found in newer units.

That's interesting. So the change didn't take place around 1995/1996, given the 1990 date code on the picture I got from your website, and I'm pretty sure that I have a fairly early (1990ish) unit that has the same hermetic 301031 part number. So you're saying these 1990-vintage VPG parts that were made well before the VHP101 series came out performed very similarly in your experience, and might have been using the same C/K foil combination? I thought they would have been K-foil parts with a significantly higher tempco than the VHP101 with its combination of C and K foil to balance the tempcos.

I had in the back of my mind that if I did a major revision on one of my 3458As that would require recalibration (like lowering the LTZ1000 temperature set point), I might as well replace R207 with an VHP101 part. But sounds like this may not yield much improvement, and may actually perform worse given the soldering heat will partially reset the aging.

[Dr. Frank]

TiN, Alm, sorry, but your statements contradict what has been found out in TiNs first 3458A repair report, credits: https://xdevs.com/fix/hp3458a/#ohm_pimp

"Most stable resistor in circuit, 40KΩ is currently specified in CLIP’s BOM as ±1.3 ppm/°C. But actual resistor on our board (and any newer 3458A’s as well), is Vishay Precision Group Foil VHP101, in hermetic oil-filled can. This series were released by VPG in 1996, so it’s later change, not reflected in old CLIP. Actual specification for VHP101 shelf-life stability is ± 0.0002% (or 2 ppm) stability in 6 years, and typical TCR ±0.3 ppm/K in +15°C to +45°C window."

The difference to the original resistor is the explicit type designator, VHP101, which is not imprinted on the old version, and obviously was not available from Vishay before 1996.
From TEMP? vs. external ohm reference measurements, I estimate that my 2001 unit with a VHP101 resistor built in has a T.C. of < 0.3ppm/K, confirming TiN own findings of VHP101 specification, and that again is in contradiction to the 1.3ppm/K in the CLIP and to the 1.1ppm/K in the 3458A specification.

Then I also want to hint to the individual serial number on top of each VHP101 type resistors, which is missing in these earlier units.
Obviously, hp ordered individually characterized (probably for T.C.?) resistors.

Frank 

The diagram shows TEMP? vs. the calculated deviation of R207 in ppm, for two different time periods, in between I made a re-calibration.
The VHP101s T.C. is specified by a window, < 10ppm in [+15 ..+ 45°C], so that's not the usual "typical" T.C. specification by Vishay.

[TiN]

I didn't say there is no difference in resistor, resistors are different at least in available specs and designators indeed. My point is that replacing old 301031 resistor with random new VHP101 will not magically improve performance as I have tested multiple 40k VHP101, VHP102 and VHP202s and they all weren't best and not usually <0.3ppm/K but larger about 0.5-1ppm/K mark. So there is little improvement to be gained replacing old 1.3ppm/K resistor with new 1ppm/K *actual* VHP101 for example.

Individual handwritten marking is a order batch number, often applied to resistors even without any special treatment/selection processes. Real PMO'd or matched resistors from VPG usually have special customer partnumber instead of generic VHP*** label on them. For example my custom VHP101 with guaranteed 0.3ppm/K:



Here's other resistor from same set



Note same part-number 321217 despite different value 

[alm]

The diagram shows TEMP? vs. the calculated deviation of R207 in ppm, for two different time periods, in between I made a re-calibration.
The VHP101s T.C. is specified by a window, < 10ppm in [+15 ..+ 45°C], so that's not the usual "typical" T.C. specification by Vishay.

How did you measure this? Hook up wires to R207 with an external 3458A or something better, and then turn on the 3458A under test and let it warm up? Or did you measure using a more indirect method?

[MegaVolt]

Here's other resistor from same set

TiN, pictures from your site are not available in Belarus writes "409 Conflict". It may be possible to load pictures on the EEVBLOG website?

[Dr. Frank]

The diagram shows TEMP? vs. the calculated deviation of R207 in ppm, for two different time periods, in between I made a re-calibration.
The VHP101s T.C. is specified by a window, < 10ppm in [+15 ..+ 45°C], so that's not the usual "typical" T.C. specification by Vishay.

How did you measure this? Hook up wires to R207 with an external 3458A or something better, and then turn on the 3458A under test and let it warm up? Or did you measure using a more indirect method?

Description of the diagram is on the bottom of my post, i.e. X-axis is TEMP? , y-axis is calculated deviation of R207 in ppm.

Latter is measured as follows:
Since 2011, I maintain a group of 10kOhm resistors, 5EA VHP202Z and later, one inside a FLUKE 5450A, which is the most stable.
I regularly measure the absolute values of all 6 resistors (once or twice a month) by my 34548A and calculate the mean value of the group, which is then my current reference value.
Btw.: These transfer measurements are uncertain to 0.2ppm StD, meanwhile, as specified in the HFL option.
This mean value is compared against calibrated references from time to time, e.g. on the Metrology Meetings, or by ab-precision, or by PTB on the Hannover Maker-Faire.

From these 'baseline' comparisons, I estimated that the VHP202Zs all show a constant drift of about -1ppm/yr. each, which is used as a prediction parameter. The 10kOhm inside the 5450A seems to be very stable.

All measurements by the 3458A can then be corrected. The correction factor, like 0.9999991, is also used to estimate the absolute value of R207, and its drift over time. So that's an indirect measurement.
As R207 is quite stable over time, I can nearly neglect its timely drift for the T.C. estimation.

The temperature of R207 is indirectly measured, as the sensor for TEMP? measurement is sitting on A1 board as well, not far away from R207, but not directly coupled of course. That generates these big outliers from the linear regression curve.
You would have to extract the timely from the temperature drift more precisely, but that's good enough for this purpose, I think.

As I would never de-solder R207, and also would not measure either its resistance nor its temperature in situ, this rough estimation method is a nice byproduct of the tracking of my resistor group, nothing more. I also use the 3458A as a ratio instrument / comparator only, neither as a voltage, nor as a resistance reference.

Frank 

[alm]

All measurements by the 3458A can then be corrected. This correction factor, so that I can estimate the absolute value of R207, and its drift over time. So that's an indirect measurement.
As R207 is quite stable over time, I can nearly neglect its timely drift for the T.C. estimation.

Thanks for explaining it! So in brief you run ACAL, and then measure a known resistance. And any change in this measurement is predominantly caused by a change (temperature or long term drift) in R207. And from this and the natural TEMP? variations due to change in ambient temperature you calculate the TC. I'm impressed that you manage this level of accuracy with so many factors involved!

[Dr. Frank]

Hello alm,
yes, that's about the procedure. I assume, that anybody might extract this from his own monitoring data, if it's done on such a regular basis... what any volt-nuts should do, anyhow, when maintaining a bunch of Volt and Ohm references.

But I would not speak about any degree of 'accuracy' (you shouldn't use this term anyhow  ), it's really a very course estimation of value and T.C. of R207 only!

Frank

[alm]

You're right that the word accuracy should not be used in metrology. When looking at the data I have from the past two years of a 1992-vintage 3458A measuring an SR104, I estimate a temperature coefficient which is quite a lot higher: -0.99 ppm/K. But looking at the graph, I'm not quite sure if this is purely the temperature coefficient of R207 or if there's more going on.

This 3458A (my #2, #1 has a temperature-dependent ACAL OHMS issue that I need to investigate so it's relegated to measuring DCV) has been measuring the SR104 (not continuously) for slightly under two years, automatically has ACAL ran anytime TEMP2? changes by more than 1 K or every 24h, and all this data gets recorded. I have a Keithley 2000 measuring the thermistor inside the SR104. From the thermistor measurement I calculated the estimated value of the SR104 based on the data in the lid re temperature coefficient of both the thermistor and the actual resistance element. I took 10 points after every ACAL, took their mean, and calculated the PPM deviation from this calculated value of the SR104, and plotted it (with the standard error of the mean, but the error bars are smaller than the points in the plot). Then I calculated linear regression, but looking at the graph and the coefficient of determination is only 0.46, I'm not very convinced that this model represents the data well. It looks like there might be two populations and a whole bunch of outliers in the data. So these would need investigations if they are related to time, changes in the setup like cabling, warmup, or other factors. But this is my best guess for now.

[Kleinstein]

Right after ACAL the resistance measurement in the 10 K range should be mainly effected by R207. ACAL should essentially transfer the R207 defined scale to other parts that need to be stable from the ACAL time to the measurement. When waiting longer after ACAL there are more parts that can dirft: there is the x10 gain to measure in the 1 V range (with 100 µA test current) and the stability of the current source itself (divider for voltage and resistor in the current source). One would see such effects when just measuring the 10 K ref for a longer time.

The noise during the resistance measurement should also be only one part of the noise. The ACAL step from R207 to the test current source also includes some noise. So repeated ACAL may give slightly different results. The 10 K range in this respect is still good and other ranges would have more of this ACAL effect.

The data indeed look a bit strange, with quite some scattering at the same temperature reading.

[Grandchuck]

From the International Vocabulary of Metrology:

(https://www.nist.gov/system/files/documents/pml/div688/grp40/International-Vocabulary-of-Metrology.pdf)

2.13 (3.5)
measurement accuracy (1)
accuracy of measurement (1)
accuracy (1)
<classical approach> closeness of agreement between a measured quantity value and a
true quantity value of the measurand
NOTES
1 — The concept ‘measurement accuracy’ is not given a numerical value, but a measurement is said
to be more accurate when it offers a smaller measurement uncertainty. Measures of
measurement accuracy are found in ISO 5725.
2 — The term “measurement accuracy” should not be used for ‘measurement trueness' and the term
“measurement precision“ should not be used for ‘measurement accuracy’.
JCGM/WG 2 Document N318
48/127
2.14
measurement accuracy (2)
accuracy of measurement (2)
accuracy (2)
<uncertainty approach> closeness of agreement between measured quantity values that
are being attributed to the measurand