EEVblog #723 - Keysight 34470A 7.5 Digit Multimeter Teardown

[EEVblog]

Would be interesting to see what Agilent competitor have put into Keithley DMM7510.

I'm supposed to have one of those on the way as well.

[EEVblog]

From what I've heard, they have also added a temperature sensor near the front panel connectors for cold point compensation (for thermocouples). As there isn't one on the measurement board, my guess is that it is on the front panel next to the connectors. I don't suppose there is any chance you fancy taking the front panel off, is there?

I had a quick look for that but didn't see it. I can only presume it is buried away under the plastic connector housing.

[EEVblog]

I was kinda disappointed lookin g at the refernce board. The old 3458A had proper vihay foil resistors, these look like (probably still very good) thin film. Also the capacitors...

Why are people expecting the 34470A to match the metrology grade 3458A?
The 34470A is designed as a cheap-ish good value bench/system 7.5 digit meter in the TrueVolt range. It is not a replacement for the 3458A for metrology applications.

[blackdog]

Hi Dave,

Because al lot of people want 3458A specs in a nice box like the 34470A...

Kind regarts,
Blackdog

[John Coloccia]

Do we seriously need to call it "Keysight", or can we just keep calling it HP or Agilent?   

[Someone]

One glaring question I have is why did they use one of Nichicon's cheapest series for the two large filter capacitors? VR series are general purpose 85C parts which aren't long life, low ESR, or anything. Couldn't they have at least gone with some 105C rated VZ series parts since those would last longer? (Capacitor life doubles for every 10C below its rated temperature.)

These high reliability devices are designed very conservatively, a proven part that has been used in designs with a long field record would be preferred over the modern equivalent even if there were some small theoretical improvement available. Shock testing and thermal cycling can be much more destructive on the tighter built capacitors and these environmental tests are the sorts of characteristics that are built into HP/Agilent/Keysight (and Tektronix) products that dont make it to the spec sheet.

[tszaboo]

I was kinda disappointed lookin g at the refernce board. The old 3458A had proper vihay foil resistors, these look like (probably still very good) thin film. Also the capacitors...

Why are people expecting the 34470A to match the metrology grade 3458A?
The 34470A is designed as a cheap-ish good value bench/system 7.5 digit meter in the TrueVolt range. It is not a replacement for the 3458A for metrology applications.

The SMD Z-Foil  resistors cost something like 5-6 dollars each, you need 4-6 of them for the LTZ1000, that could get your stability much further as it is now. And we expect things like this because the price difference between the 65 and the 70 is big, it would have fit into the gap. Also, as you pointed out, it wasn't too expensive for the current sense on the main board, there is that hermetic resistor. Also, there are some not populated part on the reference, that is where the more stable parts will go.
I'm guessing there is a 34480A coming, and they save all the possible improvement for that one. I'm actually not blaming the engineers for that, but it looks like the specifications were written by the marketing department.

[JoeN]

Since there seem to be a surplus of very knowledgeable people in this thread concerning high stability voltage references and high stability resistors, I wanted to ask this question that is bugging me:

What is the difference between the LTZ1000ACH and the LTZ1000CH?  The specs seem to give the ACH as lead free and also seem to indicate that the CH should be run hotter, but it gives the performance of both as the same.  Is that true or are one of these "better" than the other?  The LTZ1000ACH is priced as if it is the better part but I can't figure out why.  http://www.linear.com/product/LTZ1000

As far as high stability resistors go, you mention that Vishay resistors are expensive.  Is there any possibility that LT's quad matched resistor IC could be used instead?  Not that it is all that much expensive, but damn, the temperature stability on these resistors seems almost impossibly good (0.2ppm/°C and matched).  That's a lot better than what I see from Vishay.  What do you know about it?  http://www.linear.com/product/LT5400  http://cds.linear.com/docs/en/datasheet/5400fc.pdf

Thanks for any info.

[Dr. Frank]

Why are people expecting the 34470A to match the metrology grade 3458A?
The 34470A is designed as a cheap-ish good value bench/system 7.5 digit meter in the TrueVolt range. It is not a replacement for the 3458A for metrology applications.

Hi Dave,


I simply would have expected stability parameters, which are adapted to a (worst case) 6 1/2 digits reading (at e.g. 1.201000V), to really make use of the 10 times higher resolution.

That would require 0.1ppm linearity of range. As it is 1 ppm of range instead, this 1.201000V measurement is uncertain to 10 digits already. 
Further on, a 24h stability of  < 1ppm per 24h would be required, 10ppm give additional 10 digits on a 6 1/2 digits reading.
A T.C. of 1ppm/°C (instead of 0.1ppm/°C) additionally gives 5 digits at the usual +/- 5°C interval.
That sums up to 25 digits for a 1.201000 V reading, making the additional digit absolutely useless!

Instead, all these 3 crucial parameters should have been 10 times better.

That could have been easily and cheaply achieved, if the Keysight engineers would have put a little bit more brain into the analogue design, and more test and validation time into the development.
Instead they obviously simply recycled a lot.

And the 34470A is not really a cheap instrument.


The design of the 3458A is already 27 years old, so a lot of clever improvements of its circuitry elements could have been done in the meantime, at even lower cost, to be incorporated into lower resolution instruments. (The 34401A was a spin-off from the 3458A techniques)


Although promoted as a metrology grade instrument, the 3458A in practise was mainly intended for rough industrial / military environment (high Tamb. of 55°C) and not really optimized for metrology (typ. 15...35°C specification)

It is in fact designed to 7 1/2 digits performance (at 1/10 of full scale), i.e. 0.02 ppm linearity of range, 0.15ppm / °C T.C. and 1ppm 24h stability.

But in the end, 8ppm/yr. for DCV and 10ppm/yr. for OHM are really mediocre for metrology use, but that can easily be improved by proper component selection, as the Fluke 8508A / Wavetek-Datron 1271 / 1281 demonstrate.

Such an improved level of stability I would have already expected for the "new" LTZ1000 reference in the 34470A. Both the key elements, i.e. the volt and Ohm references, already have the potential for greatly improved performance ..


As a conclusion, the 34470A is really a nice instrument, with additional statistical and graphical features, and some improved characteristics.
Without competing against the still top-of-the-notch 3458A, Keysight could have designed it better, more appropriate to 7 1/2 digits,  anyhow.

Frank

[HighVoltage]

Some clever person out there will probably design a proper reference board that is a drop-in replacement for this design, which would make it perform closer to a 3458A, [but without the nifty "artifact calibration", which is one of my favorite features of the 3458A, and why I bought a 3458A and not a Fluke 8508A].

Thanks for your great explanations.
What is the  "artifact calibration" in the 3458A and why is it better than the Fluke 8508A.
Is the  "artifact calibration" similar to the ACAL of the 34470A?

[Dr. Frank]

Some clever person out there will probably design a proper reference board that is a drop-in replacement for this design, which would make it perform closer to a 3458A, [but without the nifty "artifact calibration", which is one of my favorite features of the 3458A, and why I bought a 3458A and not a Fluke 8508A].

Thanks for your great explanations.
What is the  "artifact calibration" in the 3458A and why is it better than the Fluke 8508A.
Is the  "artifact calibration" similar to the ACAL of the 34470A?

The 3458A requires a 10V and a 10k reference only, to traceably calibrate nearly ALL ranges and modes. Its ultra linear (0.02ppm of range) A/D converter is the key element for that feature, unrivaled up-to-date. See hp journal 4/1989.


All other instruments, including the 8508A / 1281 require individual Cardinal Points for each  range and each mode, instead. That sums up to about 40 different calibration values to be supplied to the instrument by a calibrator (Fluke 5720A) plus appropriate transfer standards, for correct TUR.

The 8508A / 1281 makes intensive use of Vishay BMF resistors, giving lower T.C. and higher timely  stability than in the 3458A, and also got a more stable voltage reference, an LTZ1000 or an LTFLU at about 50°C , so most of the 90 days / 1 year stability parameters are better to a factor of 2-3 over the 3458A!

The 1kV DCV range in the 8508A features the usual power dissipation compensation, which the 3458A is lacking, that leads to mediocre 12ppm uncertainty at 1000V, compared to 2ppm for the 8508A.


By simply replacing these crucial components  in the 3458A by better and actual ones, it would easily outperform the 8508A by far.

Frank

[JohnnyBerg]

I'm guessing there is a 34480A coming, and they save all the possible improvement for that one. I'm actually not blaming the engineers for that, but it looks like the specifications were written by the marketing department.

Isn't that always so?

I remember my time at the Philips lab. Always a huge battle between the commercial  and technical guys. The commercial guys wrote the specs and defined the knobs and other trivialities, and we technicians were "forced" to implemented that.

People want bits, digits and megapixels. Only a few look at the specs behind that, and realize that more bits, digits, megapixels does not always improve quality. But who cares, we got a 20 megapixel camera and a 24 bit ADC.

[HighVoltage]

The 3458A requires a 10V and a 10k reference only, to traceably calibrate nearly ALL ranges and modes. Its ultra linear (0.02ppm of range) A/D converter is the key element for that feature, unrivaled up-to-date. See hp journal 4/1989.


All other instruments, including the 8508A / 1281 require individual Cardinal Points for each  range and each mode, instead. That sums up to about 40 different calibration values to be supplied to the instrument by a calibrator (Fluke 5720A) plus appropriate transfer standards, for correct TUR.

The 8508A / 1281 makes intensive use of Vishay BMF resistors, giving lower T.C. and higher timely  stability than in the 3458A, and also got a more stable voltage reference, an LTZ1000 or an LTFLU at about 50°C , so most of the 90 days / 1 year stability parameters are better to a factor of 2-3 over the 3458A!

The 1kV DCV range in the 8508A features the usual power dissipation compensation, which the 3458A is lacking, that leads to mediocre 12ppm uncertainty at 1000V, compared to 2ppm for the 8508A.


By simply replacing these crucial components  in the 3458A by better and actual ones, it would easily outperform the 8508A by far.

Frank

Thanks so much for this detailed information.
May be Keysight is working on a successor to the 3458A?

[Dr. Frank]

Thanks so much for this detailed information.
May be Keysight is working on a successor to the 3458A?

You're welcome!

Probably not.. it's a very limited market, and the 3458A still is one of the best performers.

Maybe there's a model upgrade only, as many of its components are obsolete since a long time, and the user and bus interfaces are outdated, and these TrueVolt graphical features may be welcome.

But Keysight would have to dig deeply into the software and hardware architecture of the 3458A, and I fear, they have got rid of many of the brains back then, so they simply lack Know How to make real improvements.
I repeat again: The 34470A is mostly a copy and paste device, no really new inventions here!

Frank

[TiN]

Given all that it leaves freedom to voltnuts use own more stable LTZ1000 ref's in 34470A or even 34465A's Just design proper adapter connection and recalibrate unit.

[schopi68]

But Keysight would have to dig deeply into the software and hardware architecture of the 3458A, and I fear, they have got rid of many of the brains back then, so they simply lack Know How to make real improvements.
I repeat again: The 34470A is mostly a copy and paste device, no really new inventions here!

Terrible to see, there is a large possibility that you're right. Following the HW-Journal article authors (instrument developers) for the 3458A i could see that:

• Wayne C. Goeke is working for Keithley in the meanwhile (latest Patent 2014).
• Ronald L. Swerlein http://www.keysight.com/upload/cmc_upload/All/Swerleins_Algorithm.pdf had his latest Agilent patent 2005.
• Stephen B. Venzke had its latest Agilent Patent 2007.
• David E. Smith seems to be working for HP (latest Patent issued 2011)

Almost every one of them is aged over 55 in the meanwhile (or even retired).

[funkyant]

I learn so much from reading you guys discuss this stuff, both in this thread and all over this forum. These multimeters are well outside the scope (pardon the pun) of my profession, but it's all very interesting none the less 

[Len]

Do we seriously need to call it "Keysight", or can we just keep calling it HP or Agilent?   

Call it "that company that makes the E3631A bench supply". That covers all their names, and will probably still be right the next time they re-brand.

[nixfu]

Why couldn't they have just spun off a company called "Agilent Instruments" or something.

[Tothwolf]

One glaring question I have is why did they use one of Nichicon's cheapest series for the two large filter capacitors? VR series are general purpose 85C parts which aren't long life, low ESR, or anything. Couldn't they have at least gone with some 105C rated VZ series parts since those would last longer? (Capacitor life doubles for every 10C below its rated temperature.)

These high reliability devices are designed very conservatively, a proven part that has been used in designs with a long field record would be preferred over the modern equivalent even if there were some small theoretical improvement available. Shock testing and thermal cycling can be much more destructive on the tighter built capacitors and these environmental tests are the sorts of characteristics that are built into HP/Agilent/Keysight (and Tektronix) products that dont make it to the spec sheet.

Therein lies the rub... Nichicon does offer some high reliability series parts, but Keysight didn't even go with a general purpose 105C rated product such as the VZ series and instead used went with the cheaper general purpose 85C VR series. Will those parts last 5 years, 10 years? Sure... 20-30 years? Probably not.

I just replaced the aluminum electrolytic capacitors in one of my Tektronix 2213 (not 'A' version) scopes and even back in 1982 Tektronix used some 105C rated parts. It contained a mixture of 85C and 105C, with most of the large can parts being 105C rated.

[tszaboo]

Anyone noticed how the diode from the reference was also missing the diode in series with the heater? Or how Dave was showing the wrong page schematic for the negative reference?

BTW I just noticed they merged the two diodes into CR1. 

[schopi68]

BTW I just noticed they merged the two diodes into CR1. 

[3roomlab]

Given all that it leaves freedom to voltnuts use own more stable LTZ1000 ref's in 34470A or even 34465A's Just design proper adapter connection and recalibrate unit.

i always hear that the units are not user caliberate-able ... or that is not true? ...

*waiting for continuation of video part 2?*

[6thimage]

Given all that it leaves freedom to voltnuts use own more stable LTZ1000 ref's in 34470A or even 34465A's Just design proper adapter connection and recalibrate unit.

i always hear that the units are not user caliberate-able ... or that is not true? ...

*waiting for continuation of video part 2?*

It depends on the user, but normally no. They are not difficult to calibrate (the calibration procedure is in the back of the manual and only takes around half an hour), but they require you have specialist equipment - such as a Fluke 5720A calibrator (around $40-50k used) that has been calibrated against a higher standard. Some users will have access to the required equipment, but for the rest of us, it is easier, cheaper and more reliable to send it into a cal lab.

[Marco]

Are the front ends in these new meters still partly discrete? (ie. discrete matched FETs as inputs on the initial differential amplifier.)