HP3458A SRAM replacement: copy or calibrate?

[3db]

I get the feeling he's not going to change his views anyway.   

[View[+]Finder]

Comments and advice were much appreciated, so if I have "thanked-you," your SCAL opinions were duly noted and need no further expansion.

Generally, opinions fell into three categories:
1) "It can't be done."
2) "It shouldn't be done."
3) "It has to be done this way." (followed by a link to software or a programmer, etc.)

Please recall my reason for the post: the choice between cloning and calibrating--and as my 3458A is working fine, the ability to calibrate after NVSRAM replacement poses fewer risks for me. (The 'cloning' would have to be done in advance, as most of the posts suggest)

Several RTFM comments--they always get my attention--sent me back to the Cal Manual for a closer read. My procedure was inferred from this paragraph from the manual:
"In the following procedure, steps 1 through 12 characterize the frequency flatness of the synthesizer and cabling configuration. The equipment setting determined from this characterization are then used in the remaining steps to precisely adjust the multimeter."

In other words, the thermal conversion and voltage measurement is required to adjust for errors coming from the synthesizer. The first printing of the 3325A was in 1978 fully 10 years before the release of the 3458A, so errors were certainly anticipated and to be managed in the cal process. Once the errors are known, the synthesizer would be adjusted to conform its RMS voltage to flatten the response across frequencies. As the cal manual continues: "you can use any value between 3V and 10V rms wherever 3V is referenced, etc." Clearly, the desired quality is not the precision of the voltage as measured by the thermal converter, but rather that the synthesizer deliver a flat response.

Even the 50-ohm resistor on the input during the calibration is needed solely to deal with the impedance mismatch between the DMM and the synthesizer. To be fair, the advantage of the thermal converter approach is that the DC side of the 3458A can be used for the critical measurements and the synthesizer spec shows much finer resolution on the voltage scale.

The express purpose was to determine whether a full calibration (with SCAL) will replace nearly all of the 253 calibration constants--certainly not as a substitute for a trip to a cal lab, but as an alternative to the cloning process described elsewhere in this forum. With that in mind and not being too concerned with perfection, my process was as follows:
1) Forget the synthesizer, thermal converters and the 50-ohm resistor; work with 50-ohm output on the source and use a 50-ohm terminator on the BNC cable to banana plug at the DMM
2) Use a modern signal generator as the source (Siglent SDG2042X)
3) Use an oscilloscope to 'characterize' the output of the sig gen in the same manner as the SCAL procedure in the Cal Manual R&S HMO1212 (16bit)
4) Apply the sine wave signal to the 3458A using the sig gen and following the SCAL procedure

Only the 100mV values needed any adjustment: the sig gen was set to 125mV to deliver 100mV. The results were better (i.e. more linear) below 1MHz when comparing the output of the sig gen with the display on the 3458A. (I'll post a table of values later.) There appears to be problem with linearity above 1MHz, but that was the case before this experiment. The DC side of the 3458A does not have any problems.

[View[+]Finder]

UPDATE: 'RATIO MEASUREMENT'
Improved method for adjusting the RMS voltage from a signal generator (instead of using the thermal converters, etc.) to conform to the values expected in the 3458A SCAL calibration procedure.

The 3458A can calculate a 'ratio' measurement between a known DC 'reference' voltage on the 'sense' side and an unknown voltage on the 'input' side. The EDC CR103 DC standard was used as the reference voltage (measured to 3.00000VDC on the 34465A) and the output from the sig gen was used as the input on the 3458A. The ratio measurement was selected on the front panel ('RATIO ON') and the ACV scale was used. With the sig gen on 2MHz, the amplitude 3VRMS on the sig gen was adjusted until the ratio was as close to 1.000 as possible. Without changing any settings, the 3458A was selected to SCAL 10 and 'enter' pressed. The steps were repeated for the 8MHz 3VRMS SCAL setting. Lower frequencies (<300kHz) were set using the 34465A as on the prior post, but could have been set using the method above.

Using the ratio measurement with the 3458A, a ratio of 1.000 against a 3VDC reference means that the DC equivalent (RMS) of the AC signal is also 3VRMS. This result can be obtained with precision equal to thermal converters (home lab, not metrology) at far less cost. The completed SCAL was tested against the "Performance Test Card, 90-Day Limits" for the HP3457A (the model under the 3458A) and passed all tests up to 1MHz, the limit of the 3457A. (30V and 300V tests were not included)

The equipment required to calibrate the 3458A:
DC (Cal 0)--a 10cm piece of copper wire 1mm-1.5mm for thermal short
DC (Cal 10)--a 10V 'reference standard' (I use Ian Johnston's PDVS2, but all is required is that it be a stable, known voltage that can be input at the prompt e.g. '10.000015'
Resistance and current (Cal 10.00077E3)--a precision 10kOhm resistor (I use a Leeds & Northrup NIST traceable, but a precision resistor from Mouser might work)
AC (SCAL)--a digital signal generator (I use a 10MHz Ru standard as 'clock-in', but a 4Hr warmup is OK too); a stable, low noise DC voltage source (3V, 1V, 100mV);  If the DC side of the 3458A is OK, that can be the 'meter' for all the DC inputs.

[Dr. Frank]

UPDATE: 'RATIO MEASUREMENT'
Improved method for adjusting the RMS voltage from a signal generator (instead of using the thermal converters, etc.) to conform to the values expected in the 3458A SCAL calibration procedure.

The 3458A can calculate a 'ratio' measurement between a known DC 'reference' voltage on the 'sense' side and an unknown voltage on the 'input' side. The EDC CR103 DC standard was used as the reference voltage (measured to 3.00000VDC on the 34465A) and the output from the sig gen was used as the input on the 3458A. The ratio measurement was selected on the front panel ('RATIO ON') and the ACV scale was used. With the sig gen on 2MHz, the amplitude 3VRMS on the sig gen was adjusted until the ratio was as close to 1.000 as possible. Without changing any settings, the 3458A was selected to SCAL 10 and 'enter' pressed. The steps were repeated for the 8MHz 3VRMS SCAL setting. Lower frequencies (<300kHz) were set using the 34465A as on the prior post, but could have been set using the method above.

Using the ratio measurement with the 3458A, a ratio of 1.000 against a 3VDC reference means that the DC equivalent (RMS) of the AC signal is also 3VRMS. This result can be obtained with precision equal to thermal converters (home lab, not metrology) at far less cost. The completed SCAL was tested against the "Performance Test Card, 90-Day Limits" for the HP3457A (the model under the 3458A) and passed all tests up to 1MHz, the limit of the 3457A. (30V and 300V tests were not included)

...

This method is a cheat only.
The 3458A will first convert the ACV to a DC reading, involving its own internal AC amplifier circuit.
As this AC amplifier is not calibrated correctly in first place, also this ratio measurement is erroneous.

Therefore, you need at first to calibrate this frequency dependency of the 3458A AC amplifier by an external AC standard for 2MHz and 8MHz, BEFORE you might use any ratio AC / DC transfer. So you can't avoid an instrument like a calibrated thermo-converter.

Btw.: The 2MHz and 8MHz specification of the 3458A  is quite mediocre, like 1.5% for 2MHz and 15% for 8..10MHz at best . So your bootstrap solution also suffers from bad accuracy, as a good TUR is required.
Therefore, the 3458A will never be able to replace a TC for AC /DC conversion due to these bad specifications..
MJTCs can have a transfer accuracy of a few ppm up to 1MHz, < 0.1% at 10MHz and < 0.8% up to 100MHz.
See Table 9.7. and 9.7e., here:
https://www.nist.gov/calibrations/voltage-measurements-calibrations#533

Frank

[View[+]Finder]

Thanks for the NIST link--I've found it better to know how calibration is done at the metrology level to understand the underlying principles and the sources of error. You are absolutely correct: this is "a cheat at best." So, being an American of the old-school, I take that as a compliment.

The objective of this thread was to find a way to avoid having to disassemble a working 3458A so that the NVSRAM's could be copied before an eventual battery failure. I'm satisfied that running all the calibration routines will put my 3458A back a condition 'as good as it was before' and that for me, is good enough.

I found your comment "The 2MHz and 8MHz specification of the 3458A  is quite mediocre . . ." most interesting: if I understand correctly, the 3458A is not so good above 2MHz, so why should I get all fussy about nailing the spec, particularly on the lower voltages? Hmmm.

The weakest link in my chain is the signal generator, in particular the granularity of the the amplitude adjustment, so the voltages are just 'close'. That is where the suggested instruments shine: the RMS voltage can be spot on. Maybe I'll go shopping on eBay . . .

[texaspyro]

Look at ratio transformers.  Basically a variac with 6-7 decades of resolution.  But these have limits on what frequencies they can handle.  I have ESI and Gertsch ratio transformers.

[texaspyro]

The objective of this thread was to find a way to avoid having to disassemble a working 3458A so that the NVSRAM's could be copied before an eventual battery failure.

The calram data can be dumped over GPIB without opening the case.   See my HP3458 program in KE5FX's gpibkit package.  It will output a binary image that a device programmer can use to write a blank SRAM module.  It will also output a formatted ASCII listing of all the cal constant values.  It can also dump the data SRAMs, but that is not particularly useful since those SRAMS are also use by the firmware as general purpose memory.

[Dr. Frank]

You are absolutely correct: this is "a cheat at best." So, being an American of the old-school, I take that as a compliment.

..

I found your comment "The 2MHz and 8MHz specification of the 3458A  is quite mediocre . . ." most interesting: if I understand correctly, the 3458A is not so good above 2MHz, so why should I get all fussy about nailing the spec, particularly on the lower voltages? Hmmm.

The weakest link in my chain is the signal generator, in particular the granularity of the the amplitude adjustment, so the voltages are just 'close'. That is where the suggested instruments shine: the RMS voltage can be spot on. Maybe I'll go shopping on eBay . . .

Well, I had in mind, that you'd cheat, or fool yourself by this bootstrap idea, as it's a self-referred, or circular measurement.
Sorry, there was no pun intended, especially not related under any aspect to the current American president 

Concerning the ACV specification, please consider, which other DMM offers that high bandwidth, up to 10MHz.
The accuracy is still at 4% up to 8MHz, and I remember many old HP TVMs , like I think the hp400E, which also measured up into that frequency region, with a similar accuracy.
So these RF specs of the 3458A are quite OK, but it really shines up to 1kHz, where it competes with the highest grade calibration instruments.

Anyhow, the ACV performance is often criticised by PMEL metrologists, for some reason...this said AC pre-amplifier could have been designed much better.

The oscillator does not need fine amplitude adjustment, any random value is accepted in these cal steps.
It only has to be stable, a few tenths of percent maybe, as a transfer is already made between different output frequencies.
I also assume, that the specification is better than these 30 year old data imply, when they had to use these in-sensitive 2mV output single thermocouple. If you use a much better MJTC, like the ones offered by NIST (SRM6001), you could proof that better accuracy to a lower degree of uncertainty.

Frank

[View[+]Finder]

Hey, thanks 'pyro' for the link to the GPIB dump of the cal data. From bitter experience I learned not to click on the big green 'DOWNLOAD' button or run executables found on the 'net. I wrote a Python script that reads from the 3458A DC, AC, etc. and can probably figure out how to get the cal data, but if you have time, a few line of source code from your program would help.

I already figured out that the cal SRAM was the only one that really needed to be preserved (or not . . .) nice to have a confirmation, thanks. The key assignments can be recreated and I don't have any stored subroutines.

Here's my code:
import sys
import time
import visa

HP3458 = visa.GpibInstrument('GPIB0::22::INSTR',timeout=20,term_chars = "")
HP3458.write('END ON;RESET')
print (HP3458.ask('END ON;ID?'))

HP3458.write("PRESET NORM;OFORMAT ASCII;INBUF ON;TARM AUTO;TRIG AUTO;NPLC 10;MFORMAT 1;MEM OFF;NDIG 8;END ON")

HP3458.clear()
HP3458.write('FUNC 2')
#HP3458.write('RATIO ON')
data = open('datafile.txt','w')
print data
for x in range(0, 200):
    data.write (HP3458.ask('END ON')+ '\n')
    #time.sleep(1) # 1 sec
HP3458.write('END ON;RESET')
data.close()

That's using Keysight's latest VISA and Python 2.7.14

Cheers,
Donal

[View[+]Finder]

Thanks Dr. F
Now I know exactly what my wife will get for a holiday present: an SRM6001 for (only) $2,000! It even looks like jewelery, probaly comes in a nice little box.

Seriously, that is a fantastic creation--I had no idea NIST was in the builder space, much less a vendor--and a step in the same direction as the work with JJ's.

LoL on our 'Clown in Chief' I now have a serious case of euro-envy . . .

MFG,
Donal

[pelule]

Latest NIST MJTC Standard is the 6002
6002a Multi-Junction Thermal Converter (Nominal input resistance 600 ? ± 15 ?)  $2,000
6002b Multi-Junction Thermal Converter (Nominal input resistance 1040 ? ± 20 ?)   $2,000
(Price for ncalibrated device)
https://www.nist.gov/sri/standard-reference-instruments/sri-6002-multi-junction-thermal-converter
Specs: https://www.nist.gov/file/376126
You may also have interest into the Fluke 972A
https://www.ebay.de/sch/i.html?_from=R40&_trksid=p2055119.m570.l1313.TR0.TRC0.H0.Xfluke+792a.TRS0&_nkw=fluke+792a&_sacat=0
BR
/PeLuLe

[texaspyro]

Hey, thanks 'pyro' for the link to the GPIB dump of the cal data. From bitter experience I learned not to click on the big green 'DOWNLOAD' button or run executables found on the 'net.

GPIBKIT is open source and includes the source code.   GPIBKIT and TIMELAB are used by thousands of people over the last decade and there have never been any reports of cyber-cooties.  You can always run the .EXEs past virustotal.com and let a couple dozen different programs scan it.

[View[+]Finder]

Good tip--I didn't know about virustotal
thanks

[Dr. Frank]

Sorry, yep it's still the SRI 6002.
Was 1/2 the current price few years ago. But you get a rough cal report with this standard. < 1MHz is uncritical, anyhow, that's DCV, nearly.

It requires some additional apparatus, like relay switch, and a true nV DMM, for low ppm uncertainty.

[View[+]Finder]

SCAL completed, here are the results:

[TiN]

Sorry, yep it's still the SRI 6002.
snip.

Yep, I'm still upset on price, regretting not to buy it when it was 998$.
Is the one on photos yours? If yes, I have some questions, as this is the one way I thought about calibrating 5790A/792A/4920M.

[View[+]Finder]

UPDATE: Built-in battery backup: How long do they last?
"If it ain't broke, don't fix it" Will Rogers (just guessing)

With a working 3458A in my lab, when is the "right time" to replace the NVSRAM? "Lifetime" (well documented elsewhere . . .) is stated as 10 years, but have a look at this from the DS1230 datasheet:

The nonvolatile static RAMs constantly monitor VCC. Should the supply voltage decay, the NV SRAMs automatically write protect themselves, all inputs become “don’t care,” and all outputs become high-impedance. As VCC falls below approximately 3.0 volts, a power switching circuit connects the lithium energy source to RAM to retain data.
and this:
Each DS1230 has a built-in switch that disconnects the lithium source until the user first applies VCC. The expected tDR is defined as accumulative time in the absence of VCC starting from the time power is first applied by the user. This parameter is assured by component selection, process control, and design. It is not measured directly during production testing.

Read closely: "in the absence of VCC." This means that no power is taken from the battery while the 3458A is powered on. (Perhaps even when it is plugged in and 'off'??)

There are all sorts of lithium batteries; the one in the DS1230 are not the consumer type, they are the "really long shelf-life" chemistry as might be found in in a device that "has to work" at some time in the future and needs zero maintenance. Like, for example, a liferaft or EPIRB. The rule there is to "replace after 10 years." No exception.

As luck would have it, I happened to salvage some "replaced" 3.6V Tadiran batteries and a 9V pack from a liferaft sometime between 2001 and 2003. I figured they would be useful someday . . . and never used them. So these were at least 10 years old in 2003 (the 'rule') and possibly 25 years old today.

I measured them with my Fluke 87 (yes, it has a cal certificate from Fluke) and got 9.033V on the liferaft pack and over 3.71V on each of the four Tadiran's. (Take that as an observation, not a recommendation.)

So, if your old HP3458A on eBay really did come from "a working lab" and it was powered on as it should have been during its service, there could be many years of life left in those NVSRAM's. Oh, but if it looks like it was left out in the weather--hey, some serious science is done in severe conditions--best be prepared to work on the A5 board.

Thanks to all,
Donal

[Scopetechniques]

So you're using a device containing an LM399 to calibrate a meter that uses an LTZ1000?
The Ian Johnston voltage reference is not accurate enough to properly calibrate a 3458A.

--Victor

[View[+]Finder]

Two years have past since my last post on the HP3458A and it is still working just fine. In the meantime I sent my 34465A meter in to Keysight for a calibration and checked the 3458A against it on it return after warmup and autocal on both. Only interested in DCV and found identical out to the display limit on the 34465A. Still the same today. I do a full cal on the 3458A (same procedure, but now with better AC and RF instruments) and the 34465 is going in for calibration next year.

I bought SRAM's for the 3458A, however I've not had to install them as the lithium battery is still keeping the memory alive during the times when the 3458A is without power or unplugged (very rare). If I have to do the SRAM's, I'll upgrade the voltage reference as well and send the meter in to Keysight for calibration. Maybe.

Couple of thoughts on meters and metrology:
There is precision and there is accuracy and then there is consistency; they are not the same. It's like "rock, paper, scissors" except precision loses to accuracy and consistency in my lab.
Unlike "new wine in old bottles," using using modern instruments to calibrate (and adjust) vintage gear makes my work easier and saves time and money.