Repair and experiments thread: HP 3458A U2

[saturnin]

I know that passage very well, guess who this "Frank" was, who posed that question.

I thought so 

Very probably, there  is no market at all for such an instrument, not to talk about the R&D cost, that will never be rewarded, even over many years.

Well, 3458A is an industry standard. It covers a wide variety of applications (exactly as it was designed, i.e. to support applications in metrology (slow, but precise measurements) and at the same time to serve as fast system multimeter in production - all in one box), so imho market is not so small. 3458A is at the end of its life (see how many its internal components become obsolete), so it really makes sense to develop a new product. With a new product, they can ask more money. Although they didn't say it explicitly, they clearly stated they have been trying to do better than that (i.e. 3458A design). If you are already investing time&money in such research/development, you really want to come with a new product otherwise you become less and less competitive...

Keysight has one big advantage. Once 3458A successor is developed, only they can state it is 1:1 replacement of the older design (as they did it with 34401A->34461A upgrade). Only Keysight can guarantee full backward compatibility. And this is really important in industry applications - nobody wants to risk a loss of production due to an error in rewritten scripts for ATE...

I asked Scott Stever directly (former PM for the 34465/470A and engineer on the 3458A), and he simply refused to make any word about plans of re-design... at that instance, I had the very same feeling.

I don't know what relationships you have with him , but I would guess he simply couldn't say anything. Actually, it is probably a business secret...

For me as a PTN manager in a big electronics company , it is really a miracle how they are still able to produce that unit with all these ancient component technologies and so many obsolete parts.

I fully agree. As I wrote, it is one of reasons to move forward...

[VintageNut]

I knew that was you. I agree with everything you state except the size of the market. It is a very large market for just that one instrument.

That's to be discussed .. let's go:

They usually sold / still sell about 2000 units per year, that's about 20M$ turnover/year.
The BOM cost is quite high, as would be a re-design, with probably improved features, also HW ones.
I guess that it needs a decade to make the retrun-on-invest.
KS nowadays is completely cost driven, whereas in the 1980ties, the Cold War financed each of such exceptional projects.

So I really doubt that..
Other estimates and opinions, please!

Frank

$20M/yr is a lot of money for one instrument no matter any of your other points. There are many instruments in production that reap less then $20M/yr

[krivx]

I knew that was you. I agree with everything you state except the size of the market. It is a very large market for just that one instrument.

That's to be discussed .. let's go:

They usually sold / still sell about 2000 units per year, that's about 20M$ turnover/year.
The BOM cost is quite high, as would be a re-design, with probably improved features, also HW ones.
I guess that it needs a decade to make the retrun-on-invest.
KS nowadays is completely cost driven, whereas in the 1980ties, the Cold War financed each of such exceptional projects.

So I really doubt that..
Other estimates and opinions, please!

Frank

$20M/yr is a lot of money for one instrument no matter any of your other points. There are many instruments in production that reap less then $20M/yr

That's turnover, not profit. We're in the dark as to how much they make on each sale, and how that compares to other instruments.

[Dr. Frank]

In Europe it can be tricky next year, because unit is not RoHS compliant, but we do not know what is not RoHS compliant.

Very probably, it's simply the solder.
In automotive industry, we had to switch over to lead-free a few years ago, and that was the main issue with RoHS; other substances like Cadmium or Cr(VI) were also banned, but I think that happened already earlier. I don't think, that they used Cd alloy solder for lower thermocouple voltages.

And it's clear, that changing to lead free solder and components requires a complete redesign and re-validation.
Some special components were not available in lead free, I guess.

Frank

[TiN]

Also recently it was discovered that online shop on Linear site does not allow you to add non-Pbfree parts into cart, including LTZ1000CH/LTZ1000ACH and others hermetic metal packages I tried. #PBF part - no problem, non-PBF - page refreshed and nothing get's added

[VintageNut]

If there was no profit, they would not sell it.

I knew that was you. I agree with everything you state except the size of the market. It is a very large market for just that one instrument.

That's to be discussed .. let's go:

They usually sold / still sell about 2000 units per year, that's about 20M$ turnover/year.
The BOM cost is quite high, as would be a re-design, with probably improved features, also HW ones.
I guess that it needs a decade to make the retrun-on-invest.
KS nowadays is completely cost driven, whereas in the 1980ties, the Cold War financed each of such exceptional projects.

So I really doubt that..
Other estimates and opinions, please!

Frank

$20M/yr is a lot of money for one instrument no matter any of your other points. There are many instruments in production that reap less then $20M/yr

That's turnover, not profit. We're in the dark as to how much they make on each sale, and how that compares to other instruments.

[mimmus78]

Lock what I found on ebay!!! It's incredible!!! Never seen one of this before, it should be an experimental/prototype unit.

[TiN]

Happy holidays to every -nut  .

I'll celebrate with good DCV results on my 3458B so far. After ADC replacement, ran SN18 test for few days, to confirm board normal function.
Then 5 days meter calibrated to 10.0000V using DC standard. CAL passed w/o issues, so time to rerun SN18 with correct gain and get some initial 10V data to look at.

CAL? 72 monitoring:

Date/time	Meter temperature	CAL? 72	CAL? 1,1	CAL? 2,1	Deviation, ppm
21.DEC.2016 09:06	35.5	983.331695E-03	39.9991607E+03	7.07037491	Reference, 0.00
21.DEC.2016 20:48	36.2	983.331710E-03	+0.0153 (+0.0153)
22.DEC.2016 08:26	36.2	983.331676E-03	-0.0346 (-0.0193)
23.DEC.2016 00:52	35.6	983.331701E-03	+0.0254 (+0.0061)
23.DEC.2016 08:30	36.0	983.331641E-03	-0.0610 (-0.0549)
24.DEC.2016 17:42	36.2	983.331664E-03	+0.0234 (-0.0315)
24.DEC.2016 18:16	36.1	983.331626E-03	-0.0386 (-0.0702)
24.DEC.2016 23:21	35.5	983.331697E-03	+0.0722 (+0.0020)
25.DEC.2016 16:45	36.6	983.331594E-03	-0.1047 (-0.1027)

So far looking good, stable and behaving. 10V reading stable within 0.1ppm over 10 minutes (transfer spec) and indeed stable better than 0.3 ppm window over >24 hours (24h is 0.55ppm spec).
Room is airconditioned, so to make test more challenging, there are temperature oscillations below 1C. Still meter shows good enough data. 

10V datalog:



Live data (may take a while to draw plot, many points

RAW 10V from 732B Data for those who feel bored during holiday...

And you see this only on EEVBlog (well, and xDevs.com - how the 16 week 2016 year 3458A A3 board looks like today..
Perhaps all new 3458A from Keysight have this version of board now.



Old gate array logic chip is now replaced with a patch PCB, with ALTERA MAX CPLD in TQFP144, some 245 bus transievers and few LDOs. Transievers are due to conversion of 3.3V CMOS levels into 5V TTL for rest of the A3. More parts get obsolete - more patch boards we see..



I cooked little comparison table of common 3458A A3 PCBA versions:

A3 PCBA Version SKU	03458-66503 Rev.A	03458-66503 Rev.C	03458-66503 Rev.D	03458-66513 Rev.A	03458-66513 Rev.B
Board photograph
First released	1988	1989	~1995	2000	2015
U180 Hybrid	Same package and footprint	Same package and footprint	Same package and footprint	Same package and footprint	Same package and footprint
Interface MCU 8051	Mask 03458-85501	Mask 03458-85501	Mask 03458-85501	Rev.2 Mask , Intel 8051	Rev.2 Mask , ATMEL 8051
U210 Gate array ASIC	Fujutsu MB651314 PGA	Fujutsu MB651314 PGA	Fujutsu MB651314 PGA	HP 1820-5770 Orbit 6968A PGA	Patch PCB 03458-26550 with ALTERA CPLD
Optical interface	HP 1005-0097/HP-1005-0096	HP 1005-0097/HP-1005-0096	HP 1005-0097/HP-1005-0096	Avago HFBR-2521Z/Avago HFBR-1521Z	Avago HFBR-2521Z/Avago HFBR-1521Z
Integrating cap	MLCC 50V	50V black plastic NP0	50V black plastic NP0	yellow coated MLCC	yellow coated MLCC
U213 position	74F112PC	HP 1820-2924 IC	Jumper pin 8 - pin 10	Jumper pin 8 - pin 10	Jumper pin 8 - pin 10
ADC U142/U405 comparator	EL2018CN DIP8 (HP 1826-1817)	EL2018CN DIP8 (HP 1826-1817)	EL2018CN DIP8 (HP 1826-1817)	Patch PCB 03458-66530 with EL2251CM	Patch PCB 03458-66530 with EL2251CM
U404 ramp opamp	Hermetic LF400C	Plastic LT1122	Plastic LT1122	Plastic LT1122	Plastic LT1122
C401,Q402	Hermetic metal can	Plastic	Plastic	Plastic	Plastic
C146,C147 capacitors	None	Present	Present	Present	Present

[mimmus78]

Nice. When I finish my honeymoon with my 3458a I will open it and check how looks like. Also my unit should have a new A3 board too.

Inviato dal mio Nexus 6P utilizzando Tapatalk

[guenthert]

If there was no profit, they would not sell it.

While I have no doubts that they make some profit on the unit, this isn't necessarily the case with large companies.  It might be of 'strategic' value and funded by their marketing department.  They will make much more profit on the many lower spec'ed meters, but want to be able to say 'yes, we can' to large customers which rather deal just with one vendor.

[MadTux]

Perhaps a slightly stupid question, but that MLCC Integrating capacitor mentioned above isn't the one that does the voltage integration in the multislope ADC process?

Just wondering, don't have a 3458A (much too expensive), but know the 3456a quite well and there they always used foil capactors if they needed accurate and stable ones (e.g. the C403  Integrating cap is a foil cap). If that is indeed the integrating cap, what are the reason for using an IMO rather poor (leakage, piezoelectric microphonics, thermal drift, dissipation factor ...) ceramic cap instead of e.g PTFE foil cap?

[TiN]

Nothing stupid in your question. You correct about purpose of integrating cap. 3456A using much slower cycle and ramp times than 3458A does, so it's integrating cap much bigger in value, and need to be stable during integration time period. 3458A's ADC however is much faster, so integrating capacitor is only 330 pF. So C0G/NP0 Type I ceramic was used for it, as it's stable and much less problematic than regular Type II ceramic dielectrics. This is all to achieve max speed 100 KSPS which 3458A's ADC capable of.

I suggest great read about 3458A design in HP Journal 1989-04, where you can learn more about this instrument.

[CalMachine]

I have an idea on how to improve upon the 3458A ADC.  This idea is in it's infancy still, and I've been doing a lot of research into the subject to perfect my understanding on how to implement my idea!  If it works out, we could see DMMs with a few orders of magnitude better specs.

[Pipelie]

I have an idea on how to improve upon the 3458A ADC.  This idea is in it's infancy still, and I've been doing a lot of research into the subject to perfect my understanding on how to implement my idea!  If it works out, we could see DMMs with a few orders of magnitude better specs.

Looking forward. 

[guenthert]

I have an idea on how to improve upon the 3458A ADC.  This idea is in it's infancy still, and I've been doing a lot of research into the subject to perfect my understanding on how to implement my idea!  If it works out, we could see DMMs with a few orders of magnitude better specs.

An improved ADC can yield higher (finer) resolution and I would be surprised, if there aren't better ADCs around than the one in the HP3458a designed about 30 years ago (surely there are ADCs with higher resolution, not sure about better linearity).  Improving precision and accuracy of the overall DMM will be quite tricky though.  Hint: ask yourself what voltage the input voltage is compared to in order to yield the reported result ...

[Echo88]

The Linearity of the 3458A is, as far as i know, still unrivaled by any other DMM. Resolution means nothing, if theres not the Linearity to back it up. Apart from the fact that it needs a JJA and really good Nullvoltmeter to even test a selfbuilt 8.5-DMM-ADC which states that its as good or better than the Multislope-ADC in the 3458A.

Massive analog knowledge is what it takes to design this stuff, not new parts with better specs than their counterparts 30 years ago.

[Kleinstein]

The ADC of the 3458 is a compromise in getting high speed readings (e.g. 100 kHz) and super high resolution / linearity from the same circuit. So it is a little strange to have it as still one of the best. Today there is little need to use the same ADC for high speed readings too, as there are superior fast ADCs available (LTC2368 and similar). So it is mainly the slow high linearity part that might need to be solved better. In my view the ADC type used in NI's DMM cards with continuous integration and a separate ADC for the residual charge would be a good bet.

Still I don't think there will be so much improvement. There is noise of the reference voltage as an important source - one would need to improve there too, to have lower noise not only for a zero reading. Also many signal sources have at least comparable noise - so the only thing to do there would be to reduce the noise bandwidth by having 2 amplifiers + ADCs and do a kind of commutating auto zero - so not missing any of the signal. So this is an obvious possible factor of 2 in speed - but at a price.

Linearity at the ppb level is rather difficult - hard to model and hard to measure. Though there is a relatively easy way to check for the two most important parts of nonlinearity. DNL via the histogram and the large range INL sampling via adding voltage sources. There just is no way one can fully test a slow 40 Bit ADC - only sample testing is possible.

The more practical path would be more a smaller, cheaper solution to comes close to the 3458 performance (at longer integration times) at lower power.

There are a few modern parts that are now available that can help to make a better ADC: one are good integrated ADC to get something like 12-16 Bit readings in the MSPS range and the other is more processing power to get a more intelligent control / adjustment. So no more limit to 14 steps up and 2 steps down. Also analog simulations are much better today.

[CalMachine]

I wasn't trying to spill my guts here... but I am not sure if my idea will even increase performance of the ADC, it's just a hunch.   What I noticed while reading the specs of a lot of long scale DMMs which specifically used integrating ADC was how poor the oscillator accuracy was.  My idea is to use a GPS disciplined double ovenized oscillator doing my timing measurements of the ramp up and ramp down phases to get a more accurate time measurement with greater resolution, as well as clock references for everything.  I am also wanting to try to implement these new 2DW232 zener references  as voltage reference.

[Echo88]

I cannot comment on the clock reference, i assume if a more stable clock reference would better the 3458A specs they would have used a OCXO (since those are available since ages). As far as i read nobody has yet determined how much those 2DW232 drift over time; im more for the concept of clever filtering the LTZ1000 if needed: http://electronicdesign.com/energy/filter-trims-ultra-precision-voltage-reference

[Andreas]

I cannot comment on the clock reference, i assume if a more stable clock reference would better the 3458A specs they would have used a OCXO (since those are available since ages). As far as i read nobody has yet determined how much those 2DW232 drift over time; im more for the concept of clever filtering the LTZ1000 if needed: http://electronicdesign.com/energy/filter-trims-ultra-precision-voltage-reference

Hello,

why should the clock stability be more critical than short term stability of the integration capacitor?
Why do you think that you can filter a reference with 1.2uVpp by a chopper OP with 1.5uVpp.

The best filter for noise will be a long integration time >=100NPLC.
Of course you have to sort out the references with popcorn noise.

And in my opinion the (symmetry of) charge injection of the integrator switches is more a issue than clock stability.

with best regards

Andreas

[Echo88]

Who said that one cant use better Chopper OPs than MAX4238, for example LTC2057? For course it all has to be simulated and built in the end to test it.

[Theboel]

Integrating an GPSDO in 3458A can cause a more trouble than benefit
I do try use GPSDO to lock a stereo generator for FM transmitter and I found there is a lot of worse result in phase noise.
if I have 3458A and I need to keep it accuracy I will do 3 thing :
1. use 4 LTZ1000 in paralel to bring the reference noise down
2. Use temperature control system to keep everything inside has a stable and low humidity especially reference and ADC (yes I know about ACAL but if I keep everything 23 degree with 1/10 degree variation there is nothing wrong ?)
3. Better PSU and filtering.
 

[CalMachine]

Integrating an GPSDO in 3458A can cause a more trouble than benefit
I do try use GPSDO to lock a stereo generator for FM transmitter and I found there is a lot of worse result in phase noise.
if I have 3458A and I need to keep it accuracy I will do 3 thing :
1. use 4 LTZ1000 in paralel to bring the reference noise down
2. Use temperature control system to keep everything inside has a stable and low humidity especially reference and ADC (yes I know about ACAL but if I keep everything 23 degree with 1/10 degree variation there is nothing wrong ?)
3. Better PSU and filtering.
 

How so?  With my 53132A I can resolve 12.5 digits with the last digit fluctuating maybe 3 or 4 with my GPS reference.  That's a 12 digit stable measurement with a 1 second gate time.  This multimeter would be designed for metrology purposes, so the super fast integrating capabilities of the 3458A ADC I could care less about.  I'm aiming for like a 1 second-ish, user selectable, integration time in my design.  If you can more precisely control and time the ramp up phase as well as measuring the ramp down phase with much much greater precision and resolution... can't you then, with better certainty, determine the input signal with simple Tu/Td ratio?  Most component discrepancies are canceled out with a dual/multi slope method.

[Theboel]

How so?  With my 53132A I can resolve 12.5 digits with the last digit fluctuating maybe 3 or 4 with my GPS reference.  That's a 12 digit stable measurement with a 1 second gate time.  This multimeter would be designed for metrology purposes, so the super fast integrating capabilities of the 3458A ADC I could care less about.  I'm aiming for like a 1 second-ish, user selectable, integration time in my design.  If you can more precisely control and time the ramp up phase as well as measuring the ramp down phase with much much greater precision and resolution... can't you then, with better certainty, determine the input signal with simple Tu/Td ratio?  Most component discrepancies are canceled out with a dual/multi slope method.
[/quote]

I think its related to phase noise caused by PLL use in my GPSDO.
if you like You can try (I assume You already has 3458A) find a suitable GPSDO with a programmable output like this
http://www.leobodnar.com/shop/index.php?main_page=product_info&products_id=234
and then see how the noise level.
in my kindergarten knowledge how DMM works I still not understand what is more important low phase noise or stability of the frequency ?
if the low phase noise better to use a low phase noise OCXO like WENZEL ULN (perhaps a custom made for 3458A frequency need)   

[CalMachine]

How so?  With my 53132A I can resolve 12.5 digits with the last digit fluctuating maybe 3 or 4 with my GPS reference.  That's a 12 digit stable measurement with a 1 second gate time.  This multimeter would be designed for metrology purposes, so the super fast integrating capabilities of the 3458A ADC I could care less about.  I'm aiming for like a 1 second-ish, user selectable, integration time in my design.  If you can more precisely control and time the ramp up phase as well as measuring the ramp down phase with much much greater precision and resolution... can't you then, with better certainty, determine the input signal with simple Tu/Td ratio?  Most component discrepancies are canceled out with a dual/multi slope method.

I think its related to phase noise caused by PLL use in my GPSDO.
if you like You can try (I assume You already has 3458A) find a suitable GPSDO with a programmable output like this
http://www.leobodnar.com/shop/index.php?main_page=product_info&products_id=234
and then see how the noise level.
in my kindergarten knowledge how DMM works I still not understand what is more important low phase noise or stability of the frequency ?
if the low phase noise better to use a low phase noise OCXO like WENZEL ULN (perhaps a custom made for 3458A frequency need)   

Ahh I gotcha.  I'll definitely look into phase noise effects.

I've got 2 3458As and a 1281, as well as 2 Trimble ThunderBolt E GPS references I can use for testing purposes.