Keysight 3458A RoHS Black Edition DMM review and evaluation

Contents

Intro

Happy Metrology Day to everyone! This article was in the works since December 2019, and I apologize to everyone for such long delay for this publication. I know many people are interested in newest Keysight 3458A RoHS refresh.

Original Hewlett-Packard DMM project “SENTRY” was developed in mid’80s and introduced to market in late 1988. This project better known as model 3458A 8½-digit DMM designed for best performance money can buy. It was further evolution of already successful 6½-digit 3456A and 3457A system DMMs and very first 8½-digit for HP. In the end 3458A came out so good, that it remains in production thru all these years and today. Obsoleted parts, closed fabs and new ecology and industry regulations implemented to reduce ecology footprint and harmful heavy metals waste dictated design updates to continue production of this great instrument in recent years. Back in 80’s environment and safety regulations were not as strict as we have them now. This means a challenge for original instrument components and design to meet current RoHS compliance requirements especially with latest version that came into effect last year.

High-speed and high accuracy applications that need DMM like 3458A continued to be in demand, so just for Keysight killing off 3458A was not an option. There are very few manufacturers remain today who actually build and sell long scale 8½-digit multimeters. Most known are Keysight, Keithley (now part of Tektronix) and Fluke.
In past years Keysight had used few workarounds to sell existing stock of older non-RoHS 3458A units, but that was not long-term solution. Market demand forced DMM group to modify and facelift well proven HP3458A design to meet all latest regulations and RoHS compliance, while keeping same decades-long proven performance level of original design.

It is still amazing to see fresh 2020 production device of same design from original 1988 release. Yes, there are small updates here and there due to components obsolescence (hello, DIP-packaged opamps and hermetic can components), but at core it is still the same design, same firmware, same functionality.

If we step back in timeframe same 32 years from the oscilloscope market perspective the difference is amazing.

We went from first digital(ish) 4-channel 100 MHz HP 54500 series scopes still with CRO display to something insane like Keysight UXR with 110 GHz real-time bandwidth or smaller brother MXR.


Image 1: HP 54501A oscilloscope released same 1989 year as HP 3458A DMM

If we extrapolate same progress from oscilloscope industry and personal computer market to multimeters we should surely have 12½-digit DMM available by now, right?


Image 2: Keysight UXR 110GHz real-time oscilloscope released in 2019. Photo courtesy of Keysight web-site .

Nope, even today we still stuck at 8½-digits and have the same 3458A and it still holds a gold standard for multimeter applications in voltage measurement domain. Why all the progress in modern ASIC design, amplifiers, digital processing, memory capacity and computational performance cannot help us to get much better DMM? New opamps, computer CAD software and analytical methods should enable at least one more digit?

Reason is simple – limitations for long-scale DMMs are not in lack of post-processing but in physical abilities and analog circuit limits of the front-end and ADCs. Adding smarts in digital domain can help a bit to squeeze an little extra resolution, but does not recreate needed missing information buried under noise floor of the front-end and reference. And resolution is only small facet of performance. Proper instrument have lots of often conflicting requirements for accuracy, long-term stability, power envelope and physical design requirements.

Of course, there are technical possibilities to get higher than 8½-digit resolution for voltage measurements but these devices continue to remain only niche scientific and metrology lab domain with cryogenic agent requirements, and not something that we can get on a bench for some reasonable price (below $20k USD that is). Rack-full of expensive equipment and specialized hardware currently is the only way to get beyond 8½-digit measurement uncertainty and stability. To be fair to oscilloscopes, above mentioned Keysight UXR oscilloscope is also specialized piece of equipment out of reach for generic engineering lab as well due to high cost.

I would say even more, we may not see a magical “9½-digit” voltmeter for at least decade or two from now, as this require some new breakthroughs on new lower-noise and better stability voltage references and precision component designs. That is why in 2020 we still do not have “model 3459A”, just like flying time-travel capable DeLoreans. As result we may need to enjoy current and legacy Keysight 3458A DMMs yet many years to come.

Lot of today’s automated setups and applications still rely on and use 3458A performance. Nearly all high-precision ADC or DAC products were tested with help of this DMM at some point in time during development or production. As result existing and new customers want to preserve investments in current 3458A infrastructure, and only need new 3458A availability and 100% compatibility to the original 3458A operation procedures and firmware/software. Thanks to this Keysight applied very conservative approach for RoHS design update which affected only minimal and necessary amount of components. New 3458A RoHS DMM keep same model number, same calibration procedures, same specifications and same level of Keysight support. That also mean no new fancy interfaces, no USB, no LCD and limited amount of internal design changes.

Some people with desires for shiny touchscreens and fancy graphical UIs and snappy LXI port communications may argue that this is bad and step backwards. However, let me ask a question. If one have design proven by decades of history and use, why introduce new unknowns and complications? It is wise to preserve that existing experience and proven golden design in voltage measurement for more than 32 years with all data trace, history, stability and knowledge around that? Granted, very few instruments on the test equipment market need such special approach. HP 3458A multimeters survived rebranding three times already, first from HP to Agilent transition, and then later from Agilent to Keysight. High value of these instruments is obvious, even if we browse old and broken 3458A’s are still fetching hefty penny on ePay, despite their age and need of repair. Good used 3458A with long stability history is like rare wine, can be also quite expensive, ignoring the age of its design.

Also special ability of 3458A to perform very fast sampling with good resolution or very high accuracy with slow/moderate speeds make it as go to tool for many semiconductor characterization rigs, automated test setups, scientific equipment setups, metrology bridges and calibration systems. It is as close to golden standard in multimeter field, as one could ever be.


Image 3: Keysight 3458A RoHS refreshed 8½-digit multimeter

In this article we will look in detail on new RoHS-complaint Keysight 3458A 8½-digit multimeter. It is exact match to original instrument in size, UI, core firmware, performance and specifications. New instrument fully RoHS compliant, has modern Keysight dark-grey exterior color theme, updated PCB assemblies with obsolete components replaced and fitted with extended memory (Option 001) standard by default out of the box. More on this later in the article.

Other than new color theme update exterior design of the instrument remain unchanged. Instrument have bright single-line dot matrix VFD with excellent viewing angles and contrast (any TFT LCD are jealous), quick function change keypad section to select DCV, ACV, 2-wire Ohm, 4-wire Ohm, DCI, ACI, frequency measurement and sampling measurement modes. After few hours of using 3458A’s layout is easy to memorize and does not require reading instruction manual to perform most of the common measurement tasks. Keysight 3458A does not have functionality to measure capacitance, inductance, or diode functions with audible signals like typical benchtop DMM. It’s main purpose is high-accuracy or high-speed measurements.

User manual have some good details about more advanced topics, such as sampling configuration, precise triggering, data handling and additional setup steps for high-sleep sampling or remote programming of the instrument. In fact 3458A front panel controls are much easier than many modern touch-screeny DMMs and benchtop instruments, thanks to very flexible hot-key customization and simple layout. There are no multi-level depth menus inside system menus or fuzzy controls with 25 different functionson single key dependency.


Image 4: Keysight 3458A RoHS refreshed 8½-digit multimeter, same firmware version as old DMM

All Keysight 3458A RoHS ship with latest REV 9,2 firmware version, same as recent year production of legacy 3458A’s. All internal menu structures, commands and functions are exactly the same. There is no way to detect remotely thru GPIB if this is new instrument or old white one.

Two keypad blocks in center allow advanced setup and give controls for things like artifact calibration, NPLC sampling setting, offset compensation function for improved resistance measurements, triggering modes and autozero. Numeric keypad allows entering values, writing simple programs and scripts on meter directly or use as macro hotkeys Fn.


Image 5-6: Keysight 3458A RoHS input low-thermal binding posts and hidden fuse for current jack protection

New 3458A equipped with same custom 5-way binding post block that can accept unshrouded 4mm banana cables, bare wires or spade lug terminations. All connectors except current input are made from special copper-tellurium alloy to ensure low thermal EMF errors when used with proper copper metrology grade cables. Vertical spacing between posts is 19 mm, so standard dual-banana jacks can be accepted without problems vertically. Horizontal spacing is bit wider – 22mm.

Zero offset calibration shorting plugs designed to uniform 19 × 19 mm spacing like on most 6½-digit and 7½-digit DMMs will not work with 3458A. But with binding posts it’s not a problem, as one can always use piece of clean copper wire between all four posts to perform zero offset calibration or relative null function.

Protection fuse rated for 1 A current is hidden inside nickel plated current jack. It is removable without tools to allow easy replacement in case of current overload event and fuse damage. Just press the jack in and twist it clock-wise half a turn to release the lock and remove jack with fuse. Fuse size is standard 5 × 20 mm; type is slow-blow.

Overall this is same design as original 3458A, but on our review sample connector nuts felt a bit wobbly, with excessive freedom. It’s no functional problem, cables and connectors are still fixed very tightly once locked in place, but left a mixed feeling compared to original 3458A. Perhaps just due the early case of first production unit with updated plastic color/style and could be easily improved as the instrument manufacturing is mature.


Image 7: Keysight 3458A RoHS refreshed 8½-digit multimeter

Left and right sides of the instrument don’t have any special features, just some vent slots towards the rear to allow air exhaust. Steel strap in soft rubber envelope allow easy transport of the instrument if needed. Fan noise is noticeable, but not annoying. It is more like a low background hum. There is no fanspeed control, fan always runs on maximum speed to avoid changing thermal gradients and thermal problems around precision circuits inside.


Image 8-9: Keysight 3458A RoHS refreshed 8½-digit multimeter

Top of the front panel plastic bezel have recessed slot to allow stacking of the 3458A’s. Bottom side have matching configuration on four feet and steel clips on front feet to allow tilted setup on the bench. These feet are very common for HP/Agilent/Keysight equipment of the last 30 years.


Image 10: Keysight 3458A RoHS refreshed 8½-digit multimeter

Rear side of the instrument has intake fan opening with dust filter and protection metal grill. There are same 5-way terminals for rear input, with same spacing and layout but rotated 90-degree CCW. Serial number sticker reveals Malaysia as country of manufacture and lists options for this particular meter. All RoHS black edition 3458A have extended memory option 001 installed as factory standard, so this option is taped out with black mylar patch.

There is external isolated trigger in and trigger output BNC ports and single type of communications interface – GPIB IEEE-488. New 3458A does not understand today’s common SCPI language. Sending typical *IDN? would only trigger “command not understood” error from 3458A. Own Keysight’s software such as BenchVue promoted to work with SCPI lab instruments also does not know how to talk to 3458A properly.

Instead of SCPI both old and new 3458A talks own language, very similar to BASIC. It also has simplified state machine and programming capabilities that provide powerful customization capabilities for integration. It is non-issue for us, since all remote programming and control at xDevs.com is done via simple open-source Python applications, running on various platforms, such as PC, Raspberry Pi Linux SBC, DE1-SoC FPGA boards, etc. We will cover some programming example later in this article as well.

DMM is powered by linear custom transformer, which is configured for 110 VAC or 220 VAC operation by switches at the back. Power consumption is fairly constant and equal about 35-38W when powered by 120VAC 60Hz mains.

Pay attention to mains voltage selection switches, located near mains IEC connector. Using incorrect voltage setting (e.g. set 120 VAC with 220 VAC mains applied) will can destroy transformer because it is directly wired to mains. New transformer cost around US$420 + tax + shipping, will be rather pricey mistake to make.


Image 11: Keysight 3458A RoHS refreshed 8½-digit multimeter

Front or rear terminals selection is done via mechanical multi-pole switch just like in old 3458A. There is no way to switch selected input from remote program, which complicates use of 3458A for comparison or ratio measurements. This is one aspect where some competitor 8½-digit DMMs still offer better functionality.

Keysight is already manufacturing updated instruments, available for purchase worldwide since December 2019. Black RoHS edition instrument serial numbers start from MY59350000. Unit that we got here for testing has serial number MY59350622. Jumping ahead, this particular instrument was manufactured around 38th week of 2019, so at the time of this review it was just 16 weeks old. Essentially fresh out of press, still in the infancy.

There are also special versions of 3458A with options, such as:

Option Description Status
3458A-001 Extended memory option Standard now for RoHS DMM
3458A-002 Improved DC reference stability 4 ppm/year Available for extra $US 1.4k
3458A-E02 Special version Special SKU
3458A-H01 Special 1000 Vrms ac maximum input voltage Available for extra $US 1.5k
3458A-H06 Unknown Unknown
3458A-H11 Special PASS/FAIL BNC output connector Contact Keysight for special order
3458A-H52 US Air Force NSN Special SKU
3458A-HFL Fluke version 3458A/HFL with 2ppm/year stability and better resistance Obsoleted
3458A-OGC Precision calibration, intended for metrology use only Standard unit with $US 1.3k special calibration
3458A-700 CIIL language option
3458A-907 Front handle kit $73 USD
3458A-908 Rack flange kit $50 USD
3458A-909 Rack flange and front handle set $111 USD

Table 1: Keysight 3458A options and versions

Special “HFL option” 03458-66529 A9 reference option with 2 ppm/year stability is obsoleted in 2019 and no longer available from Keysight.

Reference stability options

Here are details how to determine installed options in each specific unit (besides looking at option labels/stickers). Option 001 (extended memory) units will return a 1,0 when queried with an OPT? command, either from front panel or GPIB. All new meters come with this option as standard (always enabled). Option 002 (high stability 4ppm/year reference – this option can be bought by ordering kit 03458-80003, around $1400 USD extra) can be checked by checked box on rear panel or verified by label on A9 PCBA under top inguard shield. The RoHS 002 A9 assembly is labeled 03458-66549 (the A9 assembly in a standard unit is labeled 03458-66539). Even better HFL 2ppm/year stability reference option, standard option for rare Fluke/HP 3458A-HFL version is not available anymore and obsoleted for the new RoHS meters.

Option H01 (1000 Volt ACV measurement capability – Standard units are limited to 700 Volt RMS for ACV) – $1500 extra.

On old units’ part-numbers used for this option are:

  • The A1 assembly is labeled 03458-91002 (the A1 assembly in a standard unit is labeled 03458-66501 or 03458-69501).
  • The A2 assembly is labeled 03458-91000 (the A2 assembly in a standard unit is labeled 03458-66502 or 03458-69502).
  • The A10 assembly is labeled 03458-91001 (the A10 assembly in a standard unit is labeled 03458-66510).

New units most likely to bear updated different PCBA part numbers as well.

Disclaimer

Redistribution and use of this article or any images or files referenced in it, in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

  • Redistributions of article must retain the above copyright notice, this list of conditions, link to this page (/article/ks3458b/) and the following disclaimer.
  • Redistributions of files in binary form must reproduce the above copyright notice, this list of conditions, link to this page (/article/ks3458b/), and the following disclaimer in the documentation and/or other materials provided with the distribution, for example Readme file.

All information posted here is hosted just for education purposes and provided AS IS. In no event shall the author, xDevs.com site, or any other 3rd party, including Keysight Technologies be liable for any special, direct, indirect, or consequential damages or any damages whatsoever resulting from loss of use, data or profits, whether in an action of contract, negligence or other tortuous action, arising out of or in connection with the use or performance of information published here.

If you willing to contribute or add your experience regarding test equipment, repairs or provide extra information, you can do so following these simple instructions

Unit in this non-paid independent review is production demo stock, received from Keysight Technologies USA for our test. All boards are production level and are typical representation. Performance of each particular instrument may be worse or better than shown in this article. For details consult with manufacturer specifications and perform your own testing.

Manuals references

Manuals for 3458A were updated to reflect cosmetic changes for the new RoHS instrument. There are no new functions or performance metric changes compared to legacy model. One rather funny mistake that caught my eye – in User’s Guide BASIC program example listing shown to test instrument response to ID? command. Manual lists response from instrument as “Keysight 3458A” but it is not correct. New RoHS DMM is identical to old one, and still responds with classy HP3458A when queried with ID?.

Keysight 3458A Multimeter datasheet

3458A Multimeter RoHS User’s Guide, Edition 1, January 9, 2020

3458A Multimeter RoHS User’s Guide, Edition 1, December 1, 2019

3458A Multimeter Legacy User’s Guide, Edition 7

A User’s Guide to Keysight 3458A Front Panel Operation (RoHS Compliance), Edition 4, August 2019

A User’s Guide to Keysight 3458A Front Panel Operation

Assembly Level Repair Manual show some low resolution photos of inner construction. We can see that layout remained the same, with most changes implemented to A1, A3 and A5 boards.

Keysight 3458A RoHS Assembly Level Repair Manual, Edition 1, December 2019

Keysight 3458A RoHS Assembly Level Repair Manual, Edition 1, March 2020

Agilent 3458A : Quick Reference Guide, Edition 2, Dec 2000

3458A Multimeter Calibration Manual, Edition 7

3458A Multimeter Calibration Manual, Edition 6

There are no public service notes currently issued for new RoHS DMM version.

Specification

Brief specification for Keysight 3458A Opt 002 (same new and old versions) listed in table below. Standard unit without 002 option has identical specifications except worse DCV long-term stability (8 ppm/year).

Function Range
DC Voltage 100mV 1 V 10 V 100V 1000 V
Accuracy (1y), Rdg ppm + Range ppm 5+3 (0.0005%) 4+0.3 (0.0004%) 4+0.05 (0.0004%) 6+0.3 6+0.1
Temperature coefficient, ppm 1.2+1 1.2+0.1 0.5+0.01 2+0.4 2+0.04
RMS noise at NPLC100, Range ppm 0.2 0.02 0.01 0.02 0.01
AC Voltage 10mV 100mV 1V 10 V 100V 1000 V
Best accuracy (1y), Rdg ppm + Range ppm 200+110 (0.02%) 70+20 (0.007%) 200+20 (0.02%) 400+20 (0.04%)
Temperature coefficient, ppm 30+200 25+1
Bandwidth frequency (max accuracy) 1 Hz to 10MHz (40Hz to 1kHz)
Resistance 4W 10 Ω 100 Ω 1 KΩ 10 KΩ 100 KΩ 1 MΩ 10 MΩ 100 MΩ 1 GΩ
Accuracy (1y), Rdg ppm + Range ppm 15+5 12+5 (0.0012%) 10+0.5 (0.0010%) 15+2 50+10 500+10 (0.05%) 5000+10 (0.5%)
Temperature coefficient, ppm 3+1 3+1 3+0.1 3+1 20+20 100+20 1000+20
RMS noise at NPLC100, Range ppm 0.3 0.3 0.03 0.045 0.06 3.6 36
DC Current 100 nA 1 µA 10 µA 100 µA 1 mA 10 mA 100 mA 1 A
Accuracy (1y), Rdg ppm + Range ppm 30 + 400 20+40 20+10 20+8 20+5 20+5 35+5 110+10
Temperature coefficient, ppm 10+200 2+20 10+4 10+3 10+2 10+2 25+2 25+3
Shunt resistance (burden voltage) 545.2kΩ(55mV) 45.2kΩ(45mV) 5.2kΩ(55mV) 720 Ω(75mV) 100 Ω(100mV) 10 Ω(100mV) 1 Ω(250mV) 0.1 Ω (<1.5V)
RMS noise at NPLC100, Range ppm 2.5 0.25 0.025 0.025 0.025 0.025 0.025 0.025
AC Current 100 µA 1 mA 10 mA 100 mA 1 A
Accuracy (1y), Rdg ppm + Range ppm 600+300 600+200 600+200 600+200 800+200
Temperature coefficient, ppm 20

Table 2: Keysight 3458A Opt 002 key specifications

More detailed specifications can be found in Appendix A of User’s Guide

Probably most asked question in the engineering community about long-scale instruments like 7½-digit and 8½-digit DMM is “why we need all these digits, in normal engineering often 4½-digits is already enough?”. This is good question that may not have obvious answers. When it comes to hobby tinkering with some Raspberry Pi and other digital modern stuff, 8½-digit DVM would not be necessary. But pure digital and embedded programming are only part of vast electronics engineering field applications. One can buy modern sigma-delta 24-bit ADC for few dollars, it’s easy to forget challenges solved during validation, testing and manufacturing of that very same ADC.

Engineer tasked with project that require accurate measurement or sourcing some analog signals would need to quantify and measure error contributors, stability, noise and overall accuracy of the working circuit. It is not that rare to find actual 16 or more bits of resolution requirement in every day hardware and tasks, such as battery analysis, power metering, sensor digitization, digital compensation for non-linear systems, physical experiments, thermocouple readouts, amplifier testing. Typically such designs can be tested by common benchtop 6½-digit DMMs, that are available from number of manufacturers which fit different budgets and feature sets. Businesses and start-ups may also choose to send their fleet of 6½-digit DMMs and sources to external 3rd party calibration house to maintain traceability and ensure correct measurements by lab instruments. However if company have large enough fleet of equipment (say tens of meters) it can quickly become significant expensive down-time. In such situation access to much more stable in-house Keysight 3458A can help to reduce costs, by acting as calibration standard.

Even if we have reasonable $100 USD per meter calibration/service fee, getting 25 DMMs done would cost good $2500 every year. And often we also have oscilloscopes, AC and DC sources, current transducers, various lab equipment (which can have also custom and complex labor-intensive calibration procedures) so total lab maintenance bill could be easily higher than cost of the few Keysight 3458A and salary of the calibration technician. After some point it can get financially cheaper to invest into metrology grade DMM like Keysight 3458A to use as reference standard, to which other every-day use equipment can be verified. This also saving lot of trouble for logistics and down-time periods during calibration, plus allows more frequent checks when needed.

Outside of calibrations, some demanding development tasks, such as validation of the high-resolution 24/32-bit ADC, 20 bit DACs or precision sensors may get requirements outside of any 6½-digit DMM capabilities. Some methods like differential measurement, using thermal chambers can help to workaround such test challenges, but for other conditions there is no replacement of the instrument that can provide much better stability, linearity and magnitude lower noise than device under test. That is what 3458A and similar grade meters are often used for as well. Even without frequent calibrations good 3458As have decades long history of stability. Thanks to high linearity and excellent noise performance it become possible to detect and characterize even smallest DUT performance aspects. After all engineering time is the most valuable resource, and best spent to work towards project goals, rather than finding workarounds of the otherwise inadequate lab instrumentation limitations.

Numerous research programs were performed to verify and characterize HP/Agilent/Keysight 3458A instruments, both in industry and in metrology fields. Some of this work is available here:

In fact, if reader have interest in use of the Keysight 3458A for digitizing and testing AC signals I can highly recommend book by R. Lapuh, ‘Sampling with 3458A’, ISBN 978-961-94476-0-4 It is reasonably priced for amount of interesting details and benchmarks presented. It reveals testing procedures and methods and typical examples where 3458A are used in metrology field by national measurement laboratories. Even if you don’t own 3458A, it is excellent source of information and ideas about signal sampling and processing.

Transfer comparison to other 8½-digit DMMs

Precision instruments like Keysight 3458A also often used in ratio mode to transfer known reference standard artifacts to unknown DUT devices. This operation require only very short-term (few minutes to hours) stability, excellent linearity and low noise. Such transfers are usually performed in stable environment conditions using best possible settings. This is similar how commercial resistance bridges and comparators operate, establishing ratio between known reference resistor and unknown resistor. Overall uncertainty is better with comparison of the unknown signal to another known reference signal versus manufacturer specifications and uncertainty for direct measurement method. Internally DMM act as ratio measurement device anyway, using internal ADC to measure difference between internal voltage/resistance/current stable sources with unknown external DUT connected to the thermals.

To compare known voltage with unknown voltage long-scale DMM used in voltmeter mode. Few decades back complex setups with sensitive null-meters like Fluke 845A, HP 419A or Keithley 155 were used in combination with saturated chemical voltage cells (and later solid-state zener references) and precision divider like Fluke 720A. With careful time consuming labor and stable standards these setups could provide voltage measurements with resolution down to 0.1 ppm for 10V range. But with 8½-digit resolution it is often possible to replace tedious measurement setups with programmable DMM like Keysight 3458A. This application is recognized by metrology equipment manufacturers, who provide not only annual stability but also short-term transfer stability specifications for their products for such short-term voltage comparisons. Here we can look at short-term compare capability for DC Voltage and Resistance functions.

This comparison completely ignore mid/long term stability and absolute calibration accuracy of the instruments. This uncertainty is NOT a combined complete measurement result, but a contribution into uncertainty budget from DMM.


Image 12: Voltage transfer specifications comparison of various long-scale DMMs

It is interesting to take a look on these specifications between different instruments. Chart derived from official specifications, it is not a measure of particular units, that expected to meet spec or demonstrate better results. Here we can spot excellent performance of Keysight 3458A reaching amazing 0.1 ppm transfer uncertainty on base 10V range full-scale. Latest and greatest DMM from competition, Fluke 8588A (also released earlier in 2019 year) demonstrate similar performance at it’s 20V range full-scale point, but degrades to 0.15 ppm at cardinal point 10V DC.

Keysight 3458A is also excelled in performance on lower 1 V and 100 mV ranges, out of the reach from more expensive Fluke 8588A. Only better competitor for these low-voltage ranges here is long obsoleted Datron/Wavetek 1281 and specialized instruments like Keysight 34420A or Keithley 2182A (for signls below 100mV). Keithley 2002 complete the performance picture, being cheapest 8½-digit DMM on the market. Half-sized Model 2002 is still able to show very good performance especially in signals outside of 3458A x1.2 fullscale multiplier. Old and obsoleted Datron 1281 demonstrate amazing best in class performance for all DC Voltages ranges except 10V. Truly sad story to see it killed off market by Fluke’s acquisition of Wavetek/Datron measurement division.

All of these DMMs in tandem with accurately calibrated reference standards can be used to improve multi-function calibrators reproduction voltage uncertainty. From chart above we can see, that even very expensive highly stable Fluke 5720A (and modern 5730A, which has exactly same DC voltage specifications) is not a match to precision 8½-digit DMM for short-term transfers. This is why most demanding calibration tasks, such as adjustment of the 7½-digit or 8½-digit customer’s DMM require combined use of the MFC and reference DMM to adjust and check calibrator’s output.

Another look on short-term resistance transfer specification. Here we have to cheat a little, because Keysight did not add resistance 10 minute specifications into current documents and manuals. Only source of such specification is addon manual from Fluke/HP (yes, for mythic HP3458A/HFL) that provide this information.

Take this 3458A data with a grain of salt below, as it is not official specification for every standard Keysight 3458A. Perhaps in future Keysight can add this into 3458A specification? From my knowledge 3458A HFL-version was using same boards and same hardware, but was subjected to pre-selection and screening to meet Fluke requirements and improved 2 ppm/year DCV specification. This gives a reason to expect ability of reaching HFL performance by standard yet well-characterized Keysight 3458A DMM.


Image 14: Resistance transfer specifications comparison of various long-scale DMMs

Based on resistance functionality transfer spec it is interesting to note 3458A’s best performing 100 kΩ range, thanks to excellent long-term stability of the reference VPG VHP101 40 kΩ. Low resistance performance on 10 Ω level is also best in class, only beat by 1.999 Ω range of Fluke 8508A. For 10 kΩ transfers 3458A can offer uncertainty about ±0.6 ppm, which is excellent for DMM.

Not taking 3458A into account, it is quite notable to see huge difference between old but mighty Datron 1281 and it’s newer reincarnations by Fluke, such as 8508A and 8588A. Latest and shiniest 8588A with pretty display, LED-indexed 5-way binding posts and modern look unable to compete in resistance transfer specifications to much older instruments. As result in resistance transfer capability Fluke 8588A despite “metrology” grade marketing efforts compete mostly with 2.5 times less expensive compact Keithley(Tektronix) Model 2002 and not the older 8508A or Datron 1281.

In resistance metrology labs DMM alone is rarely used as accurate resistance comparing device. Different kinds of bridge systems are implemented instead, providing resistance comparing capability from microohms (for example big shunts for thousands Amps) to teraohms (high-voltage dividers and alike). Transfer uncertainty of such typical yet obsolete ESI 242D bridge system from 1971 is added on Image 14 for reference. Keysight 3458A can compete with such bridge only for 100-119 kΩ point.

Green squares represent best stability that Fluke 5720A fixed resistance outputs can provide, after using manual range adjustment to external standards. This specification is valid for 24 hours, but most calibrators can keep required stability for much longer time.

Black box performance testing

This testing protocol will be used to compare meters, assuming no knowledge about instrument internal design. Just like as black box, that comply with manufacturer specifications. This mimic majority use cases for 3458A customers, including calibration service. During calibration meters are shipped to the lab, get evaluated for functions and performance with “as received” and “as returned” measurement data. Depends on meter meeting manufacturer specifications, it will be tagged in or out of tolerance and returned to customer. If customer request adjustment, then it’s possible to obtain additional data set, “as returned after adjustment”. Thanks to Artifact Calibration feature and best of the best ADC linearity unique to 3458A all adjustments are done internally by DMM itself.

Keysight 3458A RoHS Black unit was received on January 8, 2020 in original manufacturers cartoon box, bundled with standard accessory, like power cord, test leads set and calibration certificate. Units passed the initial self-test functional verification and ACAL procedure without any issues. Instrument left powered on to warm-up for 24 hours before taking any measurements. Data readout and instrument control was done using Keysight E5810A GPIB-LAN interface bridge and Terasic DE1-SoC FPGA board, running Linaro Linux 3.18.0, compiled Mon Aug 8 17:11:41 EDT 2016 for armv7l arch.


Image 14: Keysight 3458A RoHS refreshed 8½-digit multimeter reporting CAL? 2,1 value

This was a standard unit, received it’s factory calibration in Malaysia fab on December 15th, 2019. It was standard process calibration with Fluke 5720A/5725A and HP3458A as check reference standard.

Unfortunately, I was not able to get actual measurement data points and uncertainties for this review DMM, but we can somewhat guesstimate results from 90 day specifications and Keysight Malaysia fab scope of accreditation for electrical calibrations.

People around xDevs.com are not strangers to 3458A. We use these instruments a lot for voltage and resistance calibration workloads, repair of other equipment and troubleshooting of various circuits. We also maintain fleet of our own older units, many of which kept running non-stop for many years to collect history and analyze stability.

In this review we will also look at data from older DMM, using modified HP 3458A #1 and HP 3458A #2 that will be used as reference. Both of these multimeters are adjusted and calibrated with our in-house 10V DC reference and 10 KΩ standards and running 24/7/365 without power downs.

The modification in our old 3458As involve reduced oven temperature for primary reference A9 module to improve long-term stability. Stability of both meters verified by number of external calibrations with DC Voltage standards over the period till April 2020. Both instruments demonstrated annual drift under 2 ppm per year. First unit (GPIB address 3) was repaired and put in service since January 2016. The second unit (GPIB address 2) was repaired and put in service since February 2017. Both instruments are running powered up constantly 24/7 and used as system voltmeters to perform calibration tasks and act as DCV linearity standards.


Image 15: Two modified Keysight 3458A 8½-digit DMMs

To transfer cardinal voltage, resistance and current points into arbitrary values with high stability and excellent linearity we use restored and fully optioned up multi-function high-end Fluke 5720A/03 calibrator. Few additional sources and second 5720A are also sometimes used as auxiliary units for complex experiment setups, where more than one high-stability source could be needed.


Image 16: High-performance calibration station with Fluke 5720A and Fluke 5725A

Amplifier Fluke 5725A used to generate expanded AC Voltage capability for high-voltage range verification. Calibrator rig was also adjusted and verified against our lab standards. For most demanding DC voltage and resistance stability verifications DUT Keysight 3458A RoHS was tested directly against reference standards. These standards are our own 10V project in Fluke 792A power pack or Fluke 732B DC Voltage reference. For resistance tests we use Fluke SL935 or ESI SR104.


Image 17: US Lab primary DC reference bank set, 8 x Fluke 732

Fluke 732B – 10 VDC and 1.018 VDC output reference. Industry-proven DC voltage standard, widely used by many calibration and metrology labs all around the world. Usually able to deliver stability better than ±2 ppm a year with predictable linear drift. Multiple stable calibration sources are often used at metrology labs for redundancy and cross-verification. If an ensemble of four references is holding accurate value, according to statistics, it is less likely that all four would drift same way. If any single reference in bank start showing unexpected behavior due to electrical or physical damage/stress this can be detected by comparing to other references, even before the next calibration cycle to SI Volt transfer.


Image 18: Fluke 732B and prototype Fluke SL935 standards

Prototype Fluke SL935 ovenized resistance standard , with 1 Ω and 10 KΩ output. This prototype device build by secret Fluke lab, using hermetic thin-film laser trimmed resistor networks from no less than ten (!) Fluke 5720A calibrators to obtain ultra-low temperature coefficient and stability for both output values. Unit is built around 732B reference chassis, replacing original DC Voltage reference electronics with custom resistance module. Oven operating at reduced temperature, fixed around 40 °C to reduce annual drift.


Image 18: DC Voltage reference 792X based off modified Fluke 792A Power Pack

Over last few years here at xDevs.com lot of experiments were completed with our own custom designs in search of the ultimate stability DC voltage reference. Most of these utilize best reference on the market – Analog Device (former Linear Technology) LTZ1000 and LTZ1000A Super-Zener. Some publications about these designs available here, here and here.

Prototype xDevs.com 792X FX LTZ1000A reference, assembled in Fluke 792A Power Pack. This standard provides +10 VDC nominal, with tested annual drift less than -2.7 ppm and deviation from predicted linear fit less than 0.3 ppm. Key feature of this custom reference is high output drive capability at ±20 mA with minimal loading errors, due to use of true 4-wire Kelvin connection output with active output stage, unlike Fluke 732B and alike. Drawback of such design is need of bipolar power, which Fluke 792A battery power pack does provide.


Image 18: Ultra-stable oil-filled 10000 Ω primary resistance references

ESI SR104 – resistance standard with the ultra-stable resistive element, temperature sensor, all enclosed in a sealed oil-filled metal can. Everything is embedded in nice wooden box (there is an option from IET without wooden enclosure as well). ESI, being one of the industry original experts in resistance field, invented and designed legendary SR104 10000 Ω resistance standard in 1967, considered as metrology primary level grade product by many labs. This standard is still unrivaled on the market, 52 years since. Perhaps that shows how hard it is to make improvements on high-end metrology instrumentation. For fun comparison – Intel 4004 processor (released to market 4 years after ESI SR104!) evolved into modern 64-core monsters on 7 nm semiconductor process.

Test results and performance comparison

Full verification and evaluation of the complex instrument like long-scale DMM can easily take many months and sometimes even years. We don’t have such luxury here, as the Keysight 3458A was provided for my testing for just 6 weeks. I’ve decided to focus on calibration check for all functions and ranges, noise test for all DCV ranges, tempco test on few ranges/functions, linearity tests on all DCV ranges and ACAL parameters stability. These tests should give us rough idea how new RoHS version DMM compare to good old while chassis legacy units.

Artifact calibration is very powerful ace in 3458A’s sleeve, capable to save very significant support effort and cost, while keeping DMM in tight specifications. Competitor DMMs such as Keithley 2002 or Fluke 8508A/8558A/8588A require whole calibration lab setup, worth north of $200k USD to validate and maintain all functions and ranges, while for Keysight 3458A this requirement can be reduced. Table below shows comparison on calibration adjustment requirements to support DMMs for typical lab workloads:

Keysight 3458A Fluke 8508A Fluke 8588A Keithley 2002
Traceable 10V DC Voltage standard Required Required Required Not required
Calibrator, such as Fluke 5720A Not required Required Required Required
Amplifier, such as Fluke 5725A Not required Required Required Required
High-current amplifier, Fluke 52120A Not required Not required Required Not required
Resistance standard 1 Ω Not required Required Required Not required
Resistance standard 10 kΩ Required Required Required Not required
Resistance standard 1 GΩ Required, verification Required Required Required, verification
Resistance standard 10 GΩ No 10 GΩ range Required, verification No 10 GΩ range
Capacitance standard, 1000 pF No capacitance function Required No capacitance function
10 MHz frequency standard Not required Required Required Not required
1 Hz sine source Not required Required
Wideband AC source Required Not required

Nearly all 7½-digit and 8½-digit DMMs performance rely on high quality expensive components, such as ultra-stable hermetic resistor networks, carefully trimmed circuits, advanced analog circuitry and real-time parasitic corrections, sometimes with help of digital software/firmware tricks. Using best components available on the market is traditional design approach to get the job done with drawback of very high product cost. Abovementioned Fluke 8588A currently have nearly doubled price tag at $17k+ compared to price of the new Keysight 3458A as result. Requirement to have full-range lab equipment in verifications also significantly limit number of calibration labs that can support your DMM, which mean much higher ownership cost premium over years.

Keysight 3458A on other hand does not have large amount of fancy expensive components but instead rely on excellent linearity of its ADC and few high-stability internal standards to perform complete internal self-calibration and adjustment for all its functions and ranges, except niche high-frequency 1MHz+ AC bands. To obtain traceability Keysight 3458A require only two external instruments – 10V DC Voltage standard and 10 kΩ resistance standard and common copper shorting bar to perform zero offset correction. Automated processor-controlled procedure then compare internal ADC ranges to these integrated standard outputs and perform automatic transfers with sub-ppm uncertainty. The whole procedure can be done by the user in the field, and takes roughly 20 minutes, including tea break. As result whole meter is self-adjusted and brought to its 24-hour specifications on-site by user. Keysight 3458A does not allow manual adjustments or range calibration functions by user. This significantly reduces downtime for 3458A owners and eliminates risks involved in shipping of the expensive equipment to/from calibration facility.

There are two types of available ACAL procedures. User executable internal ACAL use stored constant values for 3458A’s master LTZ1000A reference and 40 kΩ hermetic resistor output to correct temperature and time drift of various circuits and functions of the meter. No external instruments or connections are required for this procedure. Whole procedure fully automatic and takes about 15 minutes, essentially bringing DMM to specifications.

Second, external CAL adjustment procedure with external artifacts perform transfers of the known 10V and 10 kΩ sources to internal references, quite equivalent to calibration adjustment procedure in conventional DMM. Then DMM performs ACAL steps to perform corrections of all the other functions and ranges from new freshly adjusted 10V and 10kΩ ranges.

External standards must be accurate enough to meet 3458A datasheet specifications:

  • 10 VDC reference (with traceable known value, to uncertainty 2 ppm or better)
  • 10 kΩ 4-wire standard reference (with traceable known value, to uncertainty 4 ppm or better)
  • Low-thermal bare copper 4-wire short for zero correction.

This ACAL process is intended to bring the multimeter to within its 24-hour specification on all available functions and ranges. This concept of internal transfers using only a two external references quite different from traditional adjustment procedure when each range/function calibrated and adjusted individually, so it is not accepted by some labs and metrologists. However, extensive research and stability studies support the claims that fully functional 3458A can indeed meet and exceed 24-hour specifications after running ACAL. Another side benefit of ACAL is that instrument can be adjusted onsite with just portable standards such as Fluke 732B and ESI SR104 as frequently as possible, making possible to keep 3458A readout as stable as external standards are.

Quite similar approach also implemented in another widely used metrology instrument – Fluke 5700A series calibrators, including obsoleted in July 2020 Fluke 5720A and current 5730A models. ACAL functionality in F5700A and better F5720A that xDevs.com lab use internally for all precision tests is more advanced, with additional 1 Ω resistance standard use, but idea of self-calibrating device where few external references used to adjust all internal ranges/functions is exactly the same. AC Voltage and AC Current functions in 5720A calibrator also require verification by additional equipment, such as external AC Transfer Voltage Standard and AC/DC current shunt set.

Self-adjustment and self-calibration concept is not something unique and also used on some older instruments even without any active electronics, such as Fluke 752A Reference Divider, Fluke 720A Kelvin-Varley Divider. Many resistance and RLC-bridges also use these self-correction concepts. Somewhat simpler ACAL is also available in modern benchtop DMM like Keysight 34465A and Keysight 34470A, however performance and transfer accuracy available in those instruments is far from 3458A level, and implementation of the ACAL functionality is very different too.

Interfacing and remote control for Keysight 3458A

Before we can dig into testing and results, we need to enable remote control for automated data logging and control of the lab equipment. Some engineers maybe got disappointed that refreshed Keysight 3458A still has only old IEEE-488 GPIB interface and did not get modern LAN LXI or forgive me, Alessandro Volta, some fancy USB Type-C or WiFi wireless interface.

Personally I think this is not important for primary customers and applications where Keysight 3458A RoHS is targeted for. Yes, cost of adding GPIB interface gateway to network/host PC is higher than having some ubiquitous USB or integrated LXI connectivity like on other modern bench T&M instruments. However, many customers who have need in 8½-digit DMM most likely already have one or two GPIB interface solutions around in the lab. Connecting and starting work with new 3458A is matter of minutes for them, so it is not a problem to have only GPIB. There can be not one, but whole stack of 3458A’s and other instruments in big rack, like on automated testing semiconductor factory floors. Rugged and robust GPIB with a parallel bus topology allow such rack to be automated from single controller, without any intermediate switches, hubs or other unwanted hardware.

We at xDevs.com used few external interfaces and software methods to interface this Keysight 3458A DMM and other lab equipment. All these interfaces can be used in standard Windows PC, but that would mean your experiments (which may take days) are depending on stable uninterrupted PC operation. This is not efficient nor convenient, so we use small low-power Raspberry Pi SBC as dedicated data logging system. Also small Raspberry Pi in metal shielded enclosure does not pose the risk of RFI/EMI around sensitive experiments that 3458A can be used for. Let’s talk briefly about GPIB hardware options first.

GPIB-USB dongle National Instruments GPIB-USB-HS


Image 19: NI GPIB USB adapter dongle

There dongles are available new or used on second-hand markets like eBay. This dongle is compatible with NI software, linux-gpib open-source library:https://linux-gpib.sourceforge.io/ and some other popular GPIB-software. There is also newer faster GPIB-USB-HS+ version, but even older ones are fast enough for every possible use case with precision DMMs. Maximum throughput that 3458A require is around 200KB/s, which is well below the capability of the GPIB-USB-HS.

GPIB-USB dongle Agilent/Keysight 82357B


Image 20: Keysight GPIB USB adapter dongle

Configuration and setup for this USB interface is almost same as NI dongle, with one additinal caveat regarding initial firmware upload. Agilent dongle using Cypress micro-controller to bridge USB data interface and GPIB controller IC. This controller store only bootloader firmware and require to have its main firmware to be uploaded on each power on.

Software package fxload can be used to accomplish this task. First, install it:

root@rpi:/home/pi# sudo apt-get install fxload

Now we need actual firmware. Download it from SourceForge repository or our local mirror

wget http://linux-gpib.sourceforge.net/firmware/gpib_firmware-2008-08-10.tar.gz
tar xvzf gpib_firmware-2008-08-10.tar.gz

Now update Cypress FX2 micro-controller with firmware. To correctly operate this required to be done two times:

root@rpi:/home/pi/gpib_firmware-2008-08-10/agilent_82357a# sudo fxload -t fx2 -D /dev/bus/usb/001/004 -I ./measat_releaseX1.8.hex

Make sure your device /dev/bus/usb/001/004 match your previous lsusb address. After first upload Agilent/Keysight adapter changed it’s ID to 001/005 in our system:

root@rpi:/home/pi/gpib_firmware-2008-08-10/agilent_82357a# lsusb
Bus 001 Device 005: ID 0957:0518 Agilent Technologies, Inc. 82357B GPIB Interface
Bus 001 Device 003: ID 0424:ec00 Standard Microsystems Corp. SMSC9512/9514 Fast Ethernet Adapter
Bus 001 Device 002: ID 0424:9514 Standard Microsystems Corp.
Bus 001 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub

Repeat loading with same firmware using command and 001/005 address now.

sudo fxload -t fx2 -D /dev/bus/usb/001/005 -I ./measat_releaseX1.8.hex

Now both green ACCESS and READY LEDs on adapter should be lit, and ID of USB dongle become 0957:0718:

root@rpi:/home/pi/gpib_firmware-2008-08-10/agilent_82357a# lsusb
Bus 001 Device 006: ID 0957:0718 Agilent Technologies, Inc.
Bus 001 Device 003: ID 0424:ec00 Standard Microsystems Corp. SMSC9512/9514 Fast Ethernet Adapter
Bus 001 Device 002: ID 0424:9514 Standard Microsystems Corp.
Bus 001 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub

To avoid doing this manually each time you plug in Agilent 82357B pod into USB, you can copy firmware file measat_releaseX1.8.hex to /usr/share/usb/agilent_82357a/ and linux-gpib hotplug function will take care of firmware upload automatically on each connection.

Using GPIB-LAN LXI bridge, such as Keysight E5810A

This is the very easy and flexible approach. HP/Agilent/Keysight E5810A and modern Keysight E5810B are essentially embedded computers running simple RTOS. These boxes work as a proxy for the GPIB data over standard Ethernet network. You can connect to same E5810A Gateway from multiple hosts and is no need to setup or install any dedicated software/drivers on the host PC. Very light-weight python-vxi11 library allow direct access to GPIB thru E5810A. You do NOT need linux-gpib installed to use this GPIB access method.

Below is example on how to install and use Keysight E5810 series LAN/GPIB gateway to communicate with Keysight 3458A and any other GPIB instruments.

1. Get SVN repository for python-vxi

svn checkout https://github.com/python-ivi/python-vxi11
cd python-vxi11/trunk

Enter /home/vxi/python-vxi11/trunk directory

python setup.py install

Now python-vxi should be installed on your linux platform, such as Raspbian OS on Raspberry Pi, that we use here.

Here’s example python app to talk with VXI instrument:

import vxi11
instr =  vxi11.Instrument("192.168.1.2", "gpib0,5") # IP address of E5810A and GPIB address of instrument
instr.timeout = 30                                  # Timeout for interface to wait, seconds
print(instr.ask("ID?"))

Make sure to type “gpib0,3” in lowercase, and don’t put any spaces between the comma and the instrument address.. Number *3 here is our GPIB instrument address, configured from front panel.

Timeout setting to 30 change the GPIB I/O timeout delay, it is important for long operations when instrument might not reply in default timeout time. This is important for example for resistance measurements with NPLC100, OCOMP and DELAY 3 with HP 3458A instruments.

In case of correct installation and connection reply can be as below:

HP3458A

Now we can write Python apps to control instruments, collect results and do post-processing and analysis. First can check correct communication with the instrument with simple script below:

# Test Python app to communicate with 3458A
import vxi11                                                  # use this command for python-vxi / e5810
inst = vxi11.Instrument("192.168.1.2", "gpib0,5")             # use this command for python-vxi / e5810
inst.write('DISP MSG,"EEVBlog Forums"')
quit()

Result should match photo below


Image 21: Keysight 3458A displays custom text message

GPIB software setup on Raspberry Pi

This step is same for both NI or Keysight dongles, only difference is configuration for board_type in /usr/etc/gpib.conf file. For NI GPIB-USB-HS dongle “ni_usb_b” setting must be used in driver parameter, while Agilent/Keysight 82357B should have “agilent_83257b”.

At this point we assume that you are familiar with basics of Raspberry Pi OS setup and have monitor/keyboard or SSH setup for headless operation.

Make sure your Pi powered with good short high-current USB cable and +5 VDC power supply with at least 2 Amp , as dongles can take some decent amount of power and with cheap USB cable connected to PC port it could cause unstable operation due to excessive voltage drop.

Script to automatically install all needed libraries drivers and packages available for download on link below. It was tested with latest Raspberry Pi OS based on Debian Buster

SH script for Raspberry Pi 3 for linux-gpib install, version 1.5 public

It should be executed with administrator rights for compatibility.

First you must agree to install system update, as it setups all required packages and libraries. Then install linux-gpib and additional BME280-based environment datalogger software after initial reboot. USB GPIB dongle must be connected to Raspberry Pi port during the installation.

BME280 setup and hardware connection also shown in better detail in this guide. BOSCH BME280 sensor is used to read environmental parameters, such as ambient temperature, relative humidity and pressure. This sensor is not mandatory but highly recommended to know if experiment results and stability are corelated with environment condition changes.

To check GPIB operation with linux-gpib interface we can use ibtest – handy CLI utility that allow manual control and command transfers between RPi and GPIB device.

root@tin:/# ibtest
Do you wish to open a (d)evice or an interface (b)oard?
        (you probably want to open a device): d
enter primary gpib address for device you wish to open [0-30]: 10
trying to open pad = 10 on /dev/gpib0 ...
You can:
        w(a)it for an event
        ...
        ...
        (r)ead string
        perform (s)erial poll (device only)
        change (t)imeout on io operations
        request ser(v)ice (board only)
        (w)rite data string
: w
enter a string to send to your device: *IDN?
sending string: *IDN?

It should return with status and with iberr equal zero. If that’s what happened, we can next readout response from the instrument.

gpib status is:
ibsta = 0x2100  < END CMPL >
iberr= 0

ibcnt = 6
You can:
        w(a)it for an event
        ...
        ...
        (r)ead string
        perform (s)erial poll (device only)
        change (t)imeout on io operations
        request ser(v)ice (board only)
        (w)rite data string
: r
enter maximum number of bytes to read [1024]: 1024
trying to read 1024 bytes from device...
received string: 'FLUKE,5720A,7543315,1.4+B+*'
Number of bytes read: 28
gpib status is:
ibsta = 0x2100  < END CMPL >
iberr= 0

ibcnt = 28

This completes the test and now everything is ready to write some programms on Raspberry Pi to control and interface lab gear.

National Instruments/Keysight VISA under Windows

This is supposed to be easiest path, but also the least reliable and user-friendly. Even if you want to send few commands to 3458A and get simple data logging script running, full VISA installation is required on your host PC. Stuff with hundreds of megabytes various libraries and server applications included in VISA packages, which you do not need for simple GPIB transfers with 3458A. Standard VISA I/O connectivity also do not speak BASIC-like 3458A programming protocol, so traditional *IDN? query to 3458A (either new or old) would just trigger an error.

To make it even more confusing, 32-bit and 64-bit versions are usually not compatible, specific driver versions may be working or not working, and each big vendor have own VISA flavor. If you have software to use Keysight VISA, it may cause big conflicts if you have Tektronix or Keithley VISA on the same system. Also if you are not using Windows, VISA is even worse option for alternative *nix OS environment.

Unless you already have NI VISA environment installed and used for your project, we would recommend against using this method to do simple measurement and setups.

Demo Python application to interface Keysight 3458A

This is all great, but now how to make actual measurement and get some data results out of the meter? For this I will use Python application to connect with GPIB interface and use instrument’s command language for configuration and data transfer.

Performing comprehensive tests wouldn’t be possible without remote control of the instruments. Simple app to perform 10V measurements and store results into CSV-like file also presented in listing below:

# xDevs.com's sample Python app to perform 10V log with 3458A
# Requirements: Python2 or Python3, python-vxi11 library
# https://xdevs.com/article/hp3458a_gpib/
# https://xdevs.com/article/ks3458b/
import sys
import time

#import Gpib                                                  # use this command for linux-gpib
#inst = Gpib.Gpib(0,5, timeout=30) # 3458A GPIB Address = 5   # use this command for linux-gpib
import vxi11                                                  # use this command for python-vxi / e5810
inst = vxi11.Instrument("192.168.1.2", "gpib0,5")             # use this command for python-vxi / e5810
inst.clear()

logname = 'log_10v_3458a_nplc200_feb2020.csv'                 # Datalog filename to save results

#Setup HP 3458A
inst.write("PRESET NORM")
inst.write("OFORMAT ASCII")
inst.write("DCV 10")
inst.write("TARM HOLD")
inst.write("TRIG AUTO")
inst.write("NPLC 200")
inst.write("AZERO ON")
inst.write("LFILTER ON")
inst.write("NRDGS 1,AUTO")
inst.write("MEM OFF")
inst.write("END ALWAYS")
inst.write("NDIG 9")

cnt = 0
tread = 1
temp = 38.5
inst.write("TEMP?")
temp = float(inst.read())
reflevel = 10.0000000
ppm = 0

with open(logname, 'a') as o:
    o.write("date;hp3458a;level;temp;ppm_level;\r\n")
    o.close()

while cnt <= 16384:                                         # take 16k samples
    with open(logname, 'a') as o:
        tread = tread - 1
        if (tread == 0):
            tread = 20                                      # read internal 3458A TEMP? every 20th sample
            inst.write("TARM SGL,1")
            inst.write("TEMP?")
            temp = inst.read()
        inst.write("TARM SGL,1")                            # Trigger DMM to take measurement
        data = inst.read()                                  # Read result
        ppm = ((float(data) / reflevel)-1)*1E6              # Calculate deviation in ppm vs reflevel
        inst.write("DISP OFF, \"%3.3f ppm\"" % float(ppm))  # Display deviation in ppm from reflevel on VFD
        print (time.strftime("%d/%m/%Y-%H:%M:%S;") + ("[%8d]: %2.8f , dev %4.4f ppm, T:%3.1f" % (cnt, float(data),float(ppm),float(temp))) )
        o.write (time.strftime("%d/%m/%Y-%H:%M:%S;") + ("%16.8f;%16.8f;%3.1f;%4.3f;\r" % (float(data),reflevel,float(temp),float(ppm))))
    o.close()
    cnt+=1

Download complete Python app for 3458A 10V logging

This Python program will generate output with readings on screen and append data to CSV-file, with name defined by logname variable. Code is compatible with both Linux system, such as Raspberry Pi and Windows, when paired with python-vxi11 library. 

d:\>python ./log_10v_3458a.py
07/02/2020-21:19:26;[       0]: 9.99996875 , dev -3.1247 ppm, T:35.0
07/02/2020-21:19:33;[       1]: 9.99996850 , dev -3.1496 ppm, T:35.0
07/02/2020-21:19:40;[       2]: 9.99996882 , dev -3.1176 ppm, T:35.0
07/02/2020-21:19:46;[       3]: 9.99996907 , dev -3.0927 ppm, T:35.0
07/02/2020-21:19:53;[       4]: 9.99996897 , dev -3.1034 ppm, T:35.0
07/02/2020-21:20:00;[       5]: 9.99996930 , dev -3.0696 ppm, T:35.0
07/02/2020-21:20:06;[       6]: 9.99996923 , dev -3.0767 ppm, T:35.0
07/02/2020-21:20:13;[       7]: 9.99996932 , dev -3.0678 ppm, T:35.0
07/02/2020-21:20:20;[       8]: 9.99996943 , dev -3.0571 ppm, T:35.0
07/02/2020-21:20:27;[       9]: 9.99996913 , dev -3.0874 ppm, T:35.0
07/02/2020-21:20:33;[      10]: 9.99996886 , dev -3.1140 ppm, T:35.0
07/02/2020-21:20:40;[      11]: 9.99996911 , dev -3.0891 ppm, T:35.0
....

Example CSV-data with 10V measurements

Output file with measurement results can be loaded to usual plotting tools for further analysis and post-processing. Python have also open-source freeware libraries that can be used to perform analysis and plot output from CSV files, such as widely used matplotlib. Steps below show application that loads CSV data, calculate axis and generate pretty plot in PNG-format.

Python plotting app, to generate PNG linear graph with dual axis. Depends on matplotlib and numpy

When executed this application parse CSV data and generate pretty plot, showing measurement results in time domain linear chart representation:


Image 22: Python + matplotlib application graph output

It also generate PNG-file which can be included in calibration reports, HTML pages, web-server status pages, etc.

Python program to program custom hotkeys for easier bench use of Keysight 3458A.

3458A has programmable hot-keys mapped to numeric keypad. Very handy feature, to preprogram keys for frequently used macros and procedures, like mode/function configurations, ACAL execution, readout of various calibration data points, etc. Below is listing from the eexample to program Fluke/HP 3458HFL keypad hotkeys and our modified functions for xDevs.com lab use. It have two sections, use one desired, while keep other one commented out. 

# xDevs.com Python test GPIB app
# http://xdevs.com/
import sys
import vxi11

inst = vxi11.Instrument("192.168.1.2", "gpib0,5") # Instrument GPIB Address = 5
inst.timeout = 30
# Fluke 3458HFL HOTKEY
'''
inst.write("DEFKEY DEFAULT")
inst.write("DEFKEY 0,'SETACV SYNC;RES .002;LFILTER ON;FUNC ACV'")
inst.write("DEFKEY 1,'LFILTER OFF'")
inst.write("DEFKEY 2,'SMATH 11,1E6;MATH 13;T 3;SMATH 11;T 1'")
inst.write("DEFKEY 3,'RMATH LOWER'")
inst.write("DEFKEY 4,'RMATH MEAN'")
inst.write("DEFKEY 5,'RMATH SDEV'")
inst.write("DEFKEY 6,'RMATH UPPER'")
inst.write("DEFKEY 7,'MATH NULL'")
inst.write("DEFKEY 8,'MATH STAT'")
inst.write("DEFKEY 9,'SMATH 7,0;MATH OFF'")
'''

# xDevs.com hotkeys
inst.write("DEFKEY DEFAULT")
inst.write("DEFKEY 0,'SETACV SYNC'")
inst.write("DEFKEY 1,'OHMF 10E3;OCOMP 1;DELAY 3;DISP ON'")
inst.write("DEFKEY 2,'MATH NULL'")
inst.write("DEFKEY 3,'MATH OFF'")
inst.write("DEFKEY 4,'OHMF 10e3;OCOMP OFF;DELAY 0;DISP ON'")
inst.write("DEFKEY 5,'DCV 1;DISP ON;NPLC 20;DELAY 0'")
inst.write("DEFKEY 6,'TEMP?'")
inst.write("DEFKEY 7,'TRIG AUTO;TARM AUTO'")
inst.write("DEFKEY 8,'OCOMP OFF;DELAY 0'")
inst.write("DEFKEY 9,'DCV 10;ACAL DCV'")
print "Programming keys complete"

Now that we ready to automate tests and collect data, let's see the results.

Warm-up time and settling

Complex analog instruments with sensitive circuits and components are subject to environmental condition changes. If you think that circuit is stable and constant, you are not looking close enough. Cold or hot temperatures, excessive humidity, elevated or reduced pressures are factors that negatively affect performance. When we talking about sub-ppm (that is below 0.0001%) changes in measurands, even small temperature change from tiny SMD LED near important analog component on PCB can cause significant deviations. Demanding measurements often require very narrow temperature range for best transfer uncertainty, otherwise additional corrections must be applied. Standard temperature for electrical measurements is +23 °C, recognized by international labs. This is why nearly all voltmeters, oscilloscopes, power supplies and other electronic test equipment state this temperature in specifications and calibration reports. If user desire to use instrument outside of this sweet spot temperature, additional temperature coefficients are oftem provided in secondary specifications.

Because of this, precision instruments also have warm-up time. Keysight provide specs for the 3458A DMM that are valid after 4 hour warm-up condition after cold state power on is met. Owner can use instrument immediately, however accuracy of the measurement is degraded.

This limited test is targeted to measure how big is the error on base 10 VDC range, if the warm-up conditioning is ignored and measurements are performed right after cold power on.

Details about used equipment in this comparison:

  • 3 × 45 cm low-thermal spade lug cable set, equivalent to Fluke 5440A-7003
  • 3458A unit, Keysight 3458A/001, reduced temperature, LTZ1000CH, <2ppm/year, calibrated 6/18/2019 to SR104/732B
  • 3458B unit, Keysight 3458A/001, reduced temperature, LTZ1000ACH, <2ppm/year, modified DCI, calibrated 1/04/2017 to xDevs US Lab SR104/732B
  • 3458 RoHS unit, Keysight Demo unit as is, calibrated 12/15/2019 by Keysight Malaysia factory.
  • xDevs.com’s calkit Python application running on DE1-SoC FPGA board for automated measurements/control
  • Agilent E5810A GPIB-LAN Gateway interface


Image 23: Cold-state 10V instability due to lack of warm-up

Black color plot on the chart represent 3458A RoHS measurement data, while blue and pink plots show two older 3458A units we have running in lab 24/7 without power down, always warm. All three instruments measure same +9.9999765 ±2 ppm VDC output from xDevs.com 792X zener-based solid-state voltage standard. As we can see, initial data have +4 ppm error. After 2 hours measured voltage reading settles at level around +1.2 ppm. Click on plot to see complete dataset chart. Waiting another 4 hours allowed 3458A RoHS to resolve voltage at +1.47 ppm off the calibrated output from standard.

Calibration accuracy and mid-term stability

To verify calibration, now we can use automated xDevs.com calkit python app to run thru various points, using same calibrator as a precision source, and each of the DMMs running same test procedure as the 3458A RoHS under test. Specification period 24 hours is used as limit levels here.

Using calibrator only to verify performance of high-precision insturment like Keysight 3458A is not the best way. Uncertainty of DMM is so low, that standard Fluke 5720A or 5730A would be challenged to provide good confidence and TUR ratio. To resolve this careful characterization and frequent calibration with fixed standards is required. We verified base DC Voltage functions and 10kΩ range directly against calibrated LTZ1000A-standard with help of Fluke 752A/Keithley 155 and 10 kΩ ESI SR104 to confirm data from calibrator tests.

Details about used equipment in this comparison:

  • Fluke 5720A/03 multifunction 7½-digit calibrator, calibrated 8/17/2019 to lab standards SR104/732B/SL935-1
  • Fluke 5725A amplifier, calibrated 8/17/2019 to lab standards SR104/732B/SL935-1
  • 3 × 45 cm low-thermal spade lug cable set, equivalent to Fluke 5440A-7003
  • 3458A unit, Keysight 3458A/001, reduced temperature, LTZ1000CH, <2ppm/year, calibrated 6/18/2019 to SR104/732B
  • 3458B unit, Keysight 3458A/001, reduced temperature, LTZ1000ACH, <2ppm/year, modified DCI, calibrated 1/04/2017 to xDevs US Lab SR104/732B
  • 3458C unit, Keysight 3458A/001/002 unmodified, calibrated 6/29/2019 to SR104/732B
  • 3458 RoHS unit, Keysight Demo unit as is, calibrated 12/15/2019 by Keysight Malaysia factory. No calibration data/uncertainty available for the unit.
  • xDevs.com’s calkit Python application running on DE1-SoC FPGA board for automated measurements/control
  • Agilent E5810A GPIB-LAN Gateway interface


Image 24: Measurement setup during tests

DC Voltage function comparison


Image 25: Calibration points and results, DC Voltage function

Each function range tested in both polarity, with 19%, 100% and 110% point, except 1kV range, which was tested at 19%, 50% and 100% point. All measurements are collected and combined MFC+DMM error from ideal value is calculated. Same test and procedure repeated for each DMM individually.

All measurements are taken with NPLC 100, AZERO ON after ACAL ALL, with floating DMM guard. Calibrator passed zero and gain calibration shortly before each test run to confirm less than 24-hour error contribution. Additional Keithley 155 and Fluke 752A reference divider were used in tandem with +10V solid-state reference for calibration checks.


Image 26: Calibration points and results, DC Voltage function

Data also presented as a graphical chart. Limits are calculated as RSS of MFC 24-hour specification and DMM 24-hour specification. This is overly strict test conditions because actual ACAL procedure on both 5720A and 3458A was NOT performed within the same 24-hour but on a much longer interval. We can see that few points on specific meters are slightly over the limits, such as +19 VDC, -19VDC on 100V range for 3458A (GPIB3) DMM and +500V, +1000V and -190V, -500V, -1000V for 3458C (GPIB11) DMM. Such test highlights the importance of using multiple DMMs to improve confidence on the actual lab capability. Even if one or two DMMs demonstrate worse accuracy, other meters can act as guard band providers on problematic tests.

Resistance function comparison

The same test repeated for resistance function, using 4-wire connection, DMM’s offset compensation function. Configuration of each meter was maintained same, and 100 &Omega – 100 kΩ ranges used OCOMP 1, DELAY 0.5 setting to mitigate cabling effects issues, as recommended by Keysight 3458A documentation.


Image 27: Calibration points and results, 4-wire resistance function

Each range was tested on 19% and 100% point, with reference resistance points provided from Fluke 5720A/03 calibrator. EXTSENSE is active on a calibrator, to enable the use of true 4-wire Kelvin connection to eliminate the impact of the cable resistance. Calibrator and DMM under test connected with 45 cm length Belden 8719 low-thermal copper spade lug cable set, similar to Fluke 5440A-7003.


Image 28: Calibration points and results, 4-wire resistance function

The graph reveals 100 kΩ point slightly towards low 24-hour specification limit, perhaps some more fine-tune is required for optimal settings. Currently, each point is sampled 30 times, and median of last 5 readings is taken as the final result. Other points reveal no issues, and system error for 100 Ω – 19 kΩ points is less than ±1.9 ppm, which is quite impressive.

AC Voltage function comparison

AC Voltage test is done in a similar manner, with test points:

  • 10 mV to 750 V AC, with 1x and 3x steps.
  • Same 45 cm Belden 8719 spade-lug cable set
  • Frequency points 10 Hz, 20 Hz, 40 Hz, 100 Hz, 1 kHz, 10 kHz, 20 kHz, 50 kHz, 100 kHz, 300 kHz, 500 kHz, and 1 MHz
  • 30 V tested with frequency up to 500 kHz
  • 100 V tested with frequency up to 100 kHz
  • 300 V tested from 100 kHz to 50 kHz, Fluke 5725A booster activated
  • 750 V tested with points 100 Hz, 1 kHz, 10 kHz and 20 kHz, Fluke 5725A booster activated


Image 29: Calibration points and results, AC Voltage function

Please click on the images to see full details on each item. Unlike other graphs, ACV test error represented in the percentage of the combined 5720A+3458A 24-specification.

Largest errors are contributed for frequency bandwidth over 50 kHz.

All 3458A’s configured with LFILTER ON, SETACV SYNC, DELAY 0, NRDGS 1, APER 0.6, RES 0.00001 command set. EXTSENSE is disabled on the calibrator for all points, only HI/LO connection is used.

DC Current function comparison

Current sourced from Fluke 5720A AUX output, with EXTGUARD disabled.


Image 30: Calibration points and results, DC current function

Each range was tested on 50% and 100% point in both polarity outputs. Calibrator and DMM under test connected with 45 cm length Belden 8719 low-thermal copper spade lug cable set, similar to Fluke 5440A-7003.

This test also clearly shows weak point, using calibrator to source currents below 100 µA due to noise floor limitation on lowest 220 µA current range. Using low-current source, such as Keithley 6430 would be better option for 3458A low current range verification.

AC Current function comparison


Image 31: Calibration points and results, AC current function

AC Flatness test


Image 40: AC test software

Noise performance testing

Noise performance of DMM was measured with shorted input, using core DC Voltage function. Noise limits standard deviation performance of the measured sampled signal at the input of the instrument. Know the noise floor of the instrument in particular range and aperture speed can be helpful to manage the overall experiment requirements.

Included noise sources are components of the signal itself and DMM-introduced sources. DMM’s noise formed by quantization noise, DNL, sampling clock jitter, own DMM’s voltage reference noise, thermal voltages, airflow drafts changes and all other possible noise sources.


Image 41: Zero noise test setup


Image 42: 100mV DC measurement noise, input terminals shorted with copper wire


Image 43: 1 V DC measurement noise, input terminals shorted with copper wire


Image 44: 10 V DC measurement noise, input terminals shorted with copper wire


Image 45: 100 V DC measurement noise, input terminals shorted with copper wire


Image 46: 1 kV DC measurement noise, input terminals shorted with copper wire

Noise below 1 mHz and 12-hour stability

Short-term stability of the DMM tandem was sampled every minute against quad LTZ1000A-based low-noise xDevs.com QVR reference over 2 day period. Results are given in table and plot below.

Composite linearity

Integral nonlinearity (acronym INL) is a specification parameter related to the performance in DAC and ADC systems. In DACs sources, it is a measure of the deviation between the ideal output function and the actual output for a certain input digital code. In ADCs, it is the deviation between the ideal input signal value and the measured represented output digital code. This measurement can be performed after other major errors, like offset and gain have been determined and possibly compensated.

Nonlinearity is one of the important metrics of ADC/DAC system or measurement device. Integral non-linearity is key factor that determine accuracy of the unknown signal measurement, after all other errors are accounted and corrected for. Quality measurement devices must specify linearity for each function range, or overall linearity of the ADC subsystem. However impact of different signal parameters and measurement device settings to the INL performance is often ommited, leaving it to the user to evaluate and characterize.

DMM’s INL is often the most challenging parameter to improve, especially for long-scale instruments. As an example for DC Voltage meter, it is relatively easy to make 7½-digit resolution or even 8½-digit ADC, but it’s extremely expensive to make same ADC that is not just high resolution but also linear to ±0.1 or better for any unknown signal applied to the ADC input. Best commercial “DAC” in the world, Fluke 720A Kelvin-Varley Divider provide linearity to ±0.1 ppm only for short-term after tedious and time-consuming calibration. Keysight 3458A DMM represent this level of performance for voltmeter performance. 3458A’s ADC is designed with INL better than ±0.1 ppm, with good units offering typical ± 0.05 ppm confirmed many times against primary Quantum Voltage Standard by number of NMLs.

We don’t have readily available JVS, so INL test resort to next best thing – use high-resolution calibrator with excellent short-term stability and tandem of DMMs for reference. That’s why chapter is called “composite linearity”, since the result represent relative figure in comparison between DMMs. Linearity was measured using a Fluke 5720A/03 calibrator and four 3458A DVMs (including RoHS DUT DMM). Measurement points were spaced by 1%, leading to a total of 200 points, starting from negative full-scale point up to positive full-scale point. All four DMMs and source are automated with Python application, linked below.

Python app used to capture linearity data from 4×3458A + 5720A, 10V range

App designed for Raspberry Pi linux environment, depends on python-vxi11 library and Keysight E5810A GPIB-LAN gateway to talk with instruments. All measurement points are captured and saved into simple CSV-file.

This data is conservative figure and does not expand the fact that multiple 3458A samples were averaged together. On the other hand, we cannot assume that the linearity of the DVMs is completely independent, but actually typical between the units. Thus, the given uncertainty is for the worst-case scenario of completely similar INL of the four 3458A. In the best-case scenario, INL is fully uncorrelated. It is extremely difficult to verify linearity for DC voltmeter to better than 0.1 ppm, due to limitation of own reference short-term instability, front-end and source noise and parasitic thermal voltages.

Pay attention to vertical scales, they are not the same between each plot! All scales represent parts per million (ppm) deviation.

Let’s start by checking 100 mVDC range. Actual range allow 20% overrange, so this is accounted in the range math. Keithley 2002 shown for comparison only, as it have 120% overrange, so it is tested only half of it’s available range. On this graph and others Keysight 3458A RoHS shown in bold red line on the plots. All ranges on instruments and source were manually fixed to prevent errors from autoranging.


Image 47: INL test, DMMs on 100mV DC range, scale is ±2.5 ppm

We can see from test result that new DMM stay inside ±0.75 ppm window and pretty close to our main 3458A-2 and 3458-3 instruments and Keithley 2002. Standard old HP 3458-11 have much bigger INL errors. Will have to take a look on that unit later on. All 3458As configured as “DCV 0.10;DISP OFF;NPLC 20;DELAY 0” and calibrator output programmed with 10 second delay to allow voltage settle on each step. This particular range is quite sensitive for ambient temperature changes and cabling.


Image 48: INL test, DMMs on 1 V DC range, scale is ±0.3 ppm

Next range. Keysight 3548A RoHS DMM stay inside ±0.1 ppm window, just like older 3458A-2 and 3458-3. Keithley 2002 shown slightly worse result at ±0.15 ppm limits. HP 3458-11 also have larger INL ±0.14 ppm. All 3458As here configured as “DCV 1;DISP OFF;NPLC 50;DELAY 0” and calibrator output programmed with 6 second delay to allow voltage settle on each step.


Image 49: INL test, DMMs on 10 V DC range, scale is ±0.1 ppm

Bonus test for closer look on 20% of the scale. This could be useful when using 3458A for voltage transfers with 1:10 ratio, such as measurement of voltage cells 1.018V versus solid-state reference with 10V output, on same 10V range. Step between each point changed by 19.0908 mV. Keysight 3548A RoHS DMM stay well inside ±0.03 ppm window. My older 3458A-2 and 3458-3 are slightly better, but we are deep down the wonderland with figures like ±0.02 ppm here. HP 3458-11 still disappointing, with +0.08 to -0.04 ppm variations. All 3458As here configured as “DCV 10;DISP OFF;NPLC 20;DELAY 0” and calibrator output programmed with 10 second delay to allow voltage settle on each step.


Image 50: INL test, DMMs on 10 V DC range, scale is ±0.2 ppm

Now the main base range, which is offering best performance. Keysight 3548A RoHS DMM shows data within ±0.05 ppm window, except -0.06 ppm peak at -7 V and increased to +0.13 ppm slope above -10.4 V. My older 3458A-2 and 3458-3 show comparable data, meeting ±0.1 ppm specification. Even HP 3458-11 is ok here. All 3458As here configured as “DCV 10;DISP OFF;NPLC 50;DELAY 0” and calibrator output programmed with 10 second delay to allow voltage settle on each step.


Image 51: INL test, DMMs on 100 V DC range, scale is ±0.3 ppm

Higher voltage range. Keysight 3548A RoHS DMM demonstrate ±0.1 ppm window. My older 3458A-2 and 3458-3 are slightly better around ±0.08 ppm here. HP 3458-11 continue to upset us, with average window ±0.15 ppm and peaks up to +0.27 ppm closer to full-scale levels. All 3458As here configured as “DCV 100;DISP OFF;NPLC 50;DELAY 0” and calibrator output programmed with 30 second delay to allow voltage settle on each step.


Image 52: INL test, DMMs on 1000 V DC range, scale is ±0.5 ppm

Test from -1090 to +1090 V, except -100 to +100V levels, which cannot be sourced by Fluke 5720A on 1kV range. Keysight 3548A RoHS DMM able to resolve ±0.15 ppm window, which slight outlier above extreme -1000 V. My older 3458A-2 and 3458-3 are slightly better around ±0.08 ppm here. HP 3458-11 on 10V range was used check unit together with freshly calibrated Fluke 752A Reference divider configured for 1:100 division, so it’s INL is not available. All 3458As here configured as “DCV 1000;DISP OFF;NPLC 20;DELAY 0” and calibrator output programmed with 20 second delay to allow voltage settle on each step.

Python analysis software used for INL plots above was verified and validated using direct DMM INL measurements against intrinsic national voltage system. Test results below demonstrate that any INL levels below 0.1 ppm indeed require cryogenic Quantum Josephson Junction Voltage system. This is also what HP used back in the days to validate ADC design and performance.

To give independent example for reference, I’ve got a dataset from such system from November 2019 to validate INL of two old 3458A (one from HP era, another from Agilent) and two Fluke 8508A reference DMMs. Direct connection between DMM and Supracon AG supraVOLTcontrol system was utilized to obtain absolute linearity plots.

JVS INL Results below are from DIFFERENT legacy white 3458A’s, not the RoHS unit from this review.

Settings for each unit were exactly the same, NPLC20,DELAY 0 for tested 3458A and RESL8,FAST_OFF,FILT_OFF mode for Fluke 8508A.


Image 53: Absolute INL test, 2×3458A and 2xFluke 8508A DMMs on 10 V range, scale ±0.1 ppm

Test against JVS offer direct measurement of DMM INL. Old HP 3548A DMM (blue plot) shows great INL within +0.05 to -0.03 ppm window. Agilent 3458A in black plot is even more impressive, revealing -0.03 to +0.02 ppm maximum error! First Fluke 8508A (red plot) show comparable data except +8 to +10V range, while second Fluke DMM is not as good, with curve window below -0.1 to +0.065 ppm.


Image 54: Absolute INL test, 3458A DMM on 1V DC range, scale ±0.25 ppm

Old 3458A also was tested on 1V range against same JVS. Plot shows INL from -0.25 to +0.1 ppm max.


Image 55: Absolute INL test, 3458A DMM on 100mV DC range

Last test on 100 mV range shows maximum INL ±0.32 ppm. Quite comparable to my own measurements with calibrator. Overall setup used in these tests shown below on Image 28.


Image 56: System setup with Supracon AG cryocooled CJVS and test DMMs in rack

Temperature stability

Temperature drift was measured in ratiometric mode, by testing +10.000VDC 0.05ppm/°C TC reference in fixed temperature thermostat, and changing ambient temperature set points from +18 °C to +25 °C. The measurement cycles typically took multiple hours after overnight cooling at 18 °C.

10V DC range

https://xdevs.com/ks3458abe_lab_tc10v_test_jan2020/

1mA DC

https://xdevs.com/ks3458abe_1mA_test_feb2020/

10mA DC

https://xdevs.com/ks3458abe_dmms_10mA_tc_feb2020/

1 A DC

https://xdevs.com/ks3458abe_1A_test_feb2020/

High-voltage settling time test

Precision 8½-digit meters are often used to perform calibration of various shunts and resistive dividers. Engineer tasked with high-voltage circuit design or source validation often need a reference divider that step high voltages down to safe levels, where they can be measured by ADC and processed further. High voltage capability of the DMM is handy for such applications, where user can perform own calibration of the divider by measuring input and the output of the divider.

However measuring high voltage by DMM also constrained by own stability and thermal/power coefficients of internal divider/attenuator inside DMM. Keysight 3458A has 10 MΩ input attenuator for 100V and 1kV DC ranges. If user apply high voltages at the input such attenuator will dissipate some power and heat up. This renders higher uncertainty for immediate readings, affected more as voltages applied are higher.

These effects are known and specified in instrument datasheet as additional 12 ppm * (VINPUT / 1000)2 for all inputs over 100 V. Thus applied 1kV will degrade uncertainty by additional ±12 ppm to the base specification of 1kV DC Voltage range.

To evaluate actual error from such case I’ve supplied stable 1000 VDC from calibrator and sampled data for about 7½ minutes. Test was automated by Python application and help of GPIB control, so there is no discrepancy in timing or sample collection. RoHS 3458A under test was tested multiple times at various days. To make test more interesting, other 3458A’s are added into comparison as well:

  • xDevs.com’s reference unit, which is our restored HP 3458A that act as reference standard DMM for lab.
  • xDevs.com’s check unit, which is secondary HP 3458A DMM.
  • Two standard HP 3458A DMM.
  • One HP 3458A DMM with factory option 002.
  • One standard Agilent 3458A DMM.


Image 29: 1kV voltage settle time in seven different 3458A DMMs from different years

Plot reveals good consistency for RoHS 3458A data, with good agreement between 6 different days and across all test points. New RoHS DMM shows readings about 2.5 ppm higher than settled level and it takes about 200 seconds for measurement to settle below 0.5 ppm of stable level. This response is also mirrored by Agilent-era 3458A, which is also a stock unit.

Our reference 3458A is special in this test, with very fast settling under 60 seconds, and even initial deviation less than 0.5 ppm. Perhaps the reason for this is installed non-factory high-speed fan, helping to keep internal temperature below +34 °C. Looking at our second check unit, one can spot the line that starts from -2 ppm and settles under 0.5 ppm window after 160 seconds.

Two other old HP 3458A show data very similar to xDevs.com’s check DMM, and one HP 3458A suffers much higher deviation, starting with +8 ppm high reading and taking about 290 seconds to settle below 0.5 ppm. To be fair, even this “bad” DMM is well within manufacturer specification, but not as good majority of DMMs in this test. This also confirms experience, that 3458A are specified very conservatively, and Keysight could pushfor better specifications without major design changes, if such customer need would arise.

Calibrator validation

One may ask, how we can trust the data above, how can we know that measurements and signals sourced from our calibrator are correct. After all we could have super-stable instrument with low noise and all, but with static error exceeding the manufacturer specifications. This is problem is solved only by performing multiple external traceable calibrations to obtain chain of measurements to national-maintained SI units. There is no other way to claim absolute calibration accuracy without actually doing external comparisons.

Being just a hobby lab, in this review we are comparing 3458A RoHS unit to ACAL-calibrated Fluke 5720A, and comparing results with data acquired by same process with three older 3458A instruments we have in lab. Calibrator and our own DMMs in lab are frequently verified against accredited commercial laboratory standards to maintain specifications under 90 day interval, often even better. All data is stored in our proprietary automation software database and subjected to statistical analysis to find out stability over time. Below I present stability charts from used Fluke 5720A to confirm compliance to 24-hour specifications at the time of this review. Data is captured over a period of 408 days, including period when Keysight 3458A RoHS in my lab between January 8, 2020 and February 18, 2020.

TCAL tag and arrow points at the date of external calibration adjustment to freshly-calibrated external standards. Traceable certificates for used standards also available on links below.

xDevs.com 792× 10V DC Reference against Fluke 732B, USA Technical Maintenance Inc, March 3, 2020

xDevs.com 792× 10V DC Reference against NIST PJVS, Taiwan NML CMS ITRI, August 8, 2019

Fluke SL935 1 Ω and 10 KΩ USA Process Insturments, 28 November 2017

Fluke SL935 1 Ω and 10 KΩ USA Process Insturments, 31 May 2018

Fluke SL935 1 Ω and 10 KΩ Taiwan NML CMS ITRI, 12 August 2019

ESI SR104 10 KΩ USA Process Insturments, 17 March 2020

At first glance graph does not look so good, but actually it is very impressive. Vertical axis show percentage of the tight 24 hour 95% Fluke 5720A specification. To determine absolute shift one you need multiply published absolute accuracy specification to percent value from the plot. For example base range 11V maximum deviation math shown below.

U11VDC = 2.75 ppm * 50% = 1.375 ppm.

Even worst range, which is passively divided 220mV offers stability better than 4.5 ppm over term more than a year, thanks to proper support with external standards and good calibrator care. In fact this confirms the widely known experience from calibration labs and previous publications, showing these instruments rather conservatively specified and offering much better performance. Especially with help of modern metrology software and historical statistic analysis and some scripting.

Next test is resistance stability for every resistance output available from Fluke 5720A. Worst change was detected on August 17, 2019 for 1900 Ω resistance output, due to fresh ACAL adjustment to Fluke SL935 10000 Ω output, externally calibrated by National Taiwan Lab CMS ITRI on August 12, 2019. This change was equal -3.28 ppm. Calibrator’s 10 kΩ output change on same date was much smaller at -2.08 ppm.

Keep in mind, these are absolute errors already including all transfer effects, not calculated relative values.

1kV AC (actually 1100V and 750V) range have also additional test points for Fluke 5725A booster. Boosted was not used all the time with calibrator under test, so some data points do not have the related data. Worst errors are around ±30% on high-frequency 750V and 1100V outputs with frequencies over 20 kHz, as expected.

DC current stability is also good, test calibrator have no problem keeping stability of the current inside ±20% deviation from it’s 24-hour 95% confidence specifications. Data for 11 A range from boosted Fluke 5725A output is also tested, but it is out of the 3458A capability and not shown here for clarity.

AC current from 10 Hz to 10 kHz specifications are also very conservative, with worst ever check deviation near -10% deviation on 220 mA 10 Hz – 500 Hz output. All other ranges and frequencies have even less error, usually less than 5%.

Glass box review

Glass box test will utilize our experience and internal design knowledge of older 3458A units, to see what improvements are implemented by Keysight in refresh design. One of such tests is base voltage gain stability, also known as CAL? 72 constant. This is a test often used to verify stability and integrity of the ultra-linear A3 ADC PCBA inside 3458A, without use of any external equipment.

Few words about old 3458As and famous ADC drift

This chapter does not really apply to 3458A RoHS DMM review, but somewhat related to a question “Did Keysight improve A3 reliability versus old instruments?”. Its no simple question, and there would be no easy answer.

There is whole lot of speculation and chatter online about failures in 3458As ADC boards. Severity of this is considered on wide scale from “nothing worth to worry about” (for owners of 3458As under warranty or service agreement plan with Keysight) to “we all doomed, my expensive meter is useless” (for hobby users that though buying $4000 unit from ePay listed as “mint” was safe). But facts are somewhere in between. There is also significant bias in reports online, just from the fact that many enthusiastic engineers purchase or acquire very old instruments from auctions or garage sales, expecting excellent performance that fully calibrated and serviced 3458A capable of.

I have reached out to Keysight to comment on A3 drift cases, however no information was shared about this matter. Everything you read below are only my own conclusions based on service and repair experience of dozen old 3458A multimeters that me or my friends own, all made from 1989 till 2006.

In fact my very first 3458A acquired in 2016 from eBay had faulty ADC that didn’t even operate, giving Error 114 : multislope convergence error. Extensive 200+ hour troubleshooting of the A3 ADC assembly confirmed culprit in U180 custom HP hybrid chip. Since you cannot buy a replacement chip from Keysight, I got another old A3 PCBA from year 2000 to replace whole thing.

That board worked but shown linear drift on base 10V DC range with rate of 1µV\V per hour. Later experiments with swapping U180 chips also confirmed that source of the drift is also U180. I have published these details in my articles on xDevs back then too and people started to send me their own experiences with old HP3458A A3 stability.


Image 59: Special modified old 03458-66503 ADC board with U180 ASIC socket, to test chips without soldering

Now there are multiple angles to address 3458A drift cases. If you are a business and use 3458A for commercial purposes, such as providing traceable calibrations, mission critical measurement or process control you already know how to contact Agilent\Keysight and get instrument repaired and verified under a warranty or service agreement. How to know that your unit require service? There are two criteria available. One is 24 hour official specification. In case of 10V DC range this parameter shall not change outside of 0.55 ppm in 24 hours since last ACAL, with ambient temperature constant. You would also need stable and verified 10V DC reference such as Datron 491X or Fluke 732. There is also more elaborate way to test ADC 10V gain stability according to Agilent Service Note 18. Once you prove drift out of published spec, you can expect a repair, given current and active warranty on your unit. End of the story.

Now, the more complicated case, more in line of enthusiast community who have 3458A in the home labs, or those business who count on luck and consider paying for warranty is not worthy investment. Often these owners got 3458A from random garage\warehouse , with zero history and without knowledge about instrument storage condition, previous issues and rejection reasons. Some resellers may swap random boards between different 3458A units in hope to see “SELFTEST PASSED” to do a quick sale. I have bought four of these old unknown 3458As myself in pursuit of repair challenge and knowledge gain. Now, three times out of my own four meters I have measured excessive drift on base 10V range. You could think “this is 75% failure rate”.

But actually it is not this bad, because of different pass/fail conditions. I had desire to have not just “in-spec” 3458A but one that is much better than spec. After analyzing data from other 3458A’s from different owners I have concluded that some 3458A able to stay inside 24 hour spec for days and weeks WITHOUT mandatory daily ACAL correction. This is very important point, and it is not covered by specification from HP/Agilent/Keysight. It is easy to assume that 1 year specification provided by Keysight applies to instrument as is (after warmup time, of course) just like any other competition DMM like Keithley 2002, Fluke 8588A, Datron 1281. But its not correct, 1 year spec valid on 3458A only after ACAL procedure. And as expected this ACAL procedure corrects A3 ADC errors, including linear drift every time its executed. Hence normal use cases , when ACAL performed every day or every time before important measurement will still give decent data. So less demanding users may as well perform ACAL every time best accuracy is desired, even with drifty ADC.

There is a catch however, more related to the issue itself. Rumors are that core of the ADC, hybrid chip U180, had a transition to another production fab back in 2000s. This could affect materials used such as package epoxy bonds, bare die attach and processing. During the changed processes, some batch of U180’s had excessive stress on resistor network die and over time it caused mismatched drift in different resistors. I’ve seen before an excessive drift in film resistors caused by mechanical strain, measured separately on test resistors. This theory results also confirmed by my own tests on bad U180 chip, measuring ADC’s reference +12, -12 and +5V voltages that are scaled by U180 resistors and opamps in A3 PCBA. Drifty chips had these voltages with mismatched drift, causing errors in multislope operation, that expects stable currents per each conversion step. As a result, bad U180 has not only linear drift but also degraded linearity performance. I have seen linearity as bad as 0.6ppm on 10V range, while good fully working 3458A have linearity around 0.05 ppm. It is far less known fact, because few enthusiast have ability or desire to test linearity with sub-ppm sensitivity.


Image 59: Integral linearity test of three 3458A, with unit © having drifting A3 ADC board

INL plot from Image 59 shows 300% out of spec linearity error measured by test HP 3458A on base 10V range, suffering from bad A3 PCBA (bold green lines). 0.3 ppm error might not sound like much, but it makes large error when such an instrument used in ratio mode. So it is important to keep an eye on INL and drift to confirm good confidence in measurement results.

Another important factor is to understand how ADC and ACAL procedure operates. By design DMM considers onboard LTZ1000A as constant voltage source. So “ADC” drift that you measure with SN18 procedure is in fact calculated value between ADC gain at the time of ACAL procedure and constant value of LTZ1000A A9 module output stored in NVRAM. There is no way to measure own A9 LTZ1000A drift without external standard of similar grade. To be fair, after seeing data from more than 20 different HP and Agilent 3458As (up to units from year 2015) I did not see A9 LTZ drift more that 24 hour spec of 0.55 ppm. But some units needed multiple days of uninterrupted powered on warmup after lengthy (months) storage in cold state before they get to stable state within short-term noise specifications. Based on this statistics I considered drift of internal A9 is unlikely, but still modified my own 3458A to run at lower oven temperature and keep them powered on 24\7 whole year in a airconditioned room at +23 ±2 °C environment. This also helps in my goal to provide 3458A measurements that are better than published spec. Also if 3458A is used for ratio measurements or AC voltage sampling, drift and degraded linearity even at 0.55 ppm levels would not be significant contributors to the result.

I have not heard of anybody with Keysight 3458A that is manufactured no later than 2014 to have bad or drifty ADC board. There are some bad “refurbished” A3 boards around, but not the new assemblies. Perhaps it would be interesting to collect a database of the privately owned instruments and tested drift, even if its CAL? 72 value per SN18 procedure. If you have 3458A and willing to participate – feel free to contact us at xDevs chat or via email team[at]xdevs.com.

Now that we have ADC question covered, it is time to look inside and reveal what was updated or changed versus original instrument design.


Image 59: Keysight 9100-4715 special shielded power transformer

Main custom transformer is E-core type, mounted on sturdy metal chassis frame right behind the front panel and VFD screen, far from all critical and sensitive analog components. This transformer has dual shielding and guarding cage, to reduce mains noise coupling and provide clean isolated power domains between digital power and analog power.


Image 60: Transformer schematics

Typical to any other 8½-digit DMMs, no switch-mode power supplies are allowed in analog domain. That means using of bulky 50/60 Hz linear power transformer for all isolated power supply presented to the boards.

Switch-mode power supplies offer higher power density (reduced weight and compact size) and better power efficiency. But these aspects not as important as keeping low noise clean power in this instrument type. Switchmode supplies generate lots of unwanted ripple and noise due to high-bandwidth switching pulse nature of their operation. Special switching converters with controlled slew-rate exist, but they require extensive care in design and more expensive than good old linear regulator.

Low noise clean power is a first and often critical step in high-performance analog circuit design. Power consumption of analog circuits also relatively low, we don’t have juicy processors that may require tens of amps. So no need of switching high-efficiency converters in 3458A, good shielded linear transformer-based regulator still does the job fine.


%(imgref)Image 59: Intake ebm papst 622 L high-quality fan.

Just like original 3458A, rear fan oriented to intake cool air at the rear of the instrument and blow it thru the inguard power regulator, outguard digital controller board and exhaust on the left back side of the DMM.

Polycarbonate separator around A4 inguard power PCBA directs airflow towards A5 board and help to reduce amount of drafts going over A1, A2 and A3 analog boards. DC Fan is high-quality 12V type, manufactured by ebm papst, model 622 L. This fan is rated for 80k hours of life, has ball-bearing suspension and produce audible but not annoying level of low-frequency noise.


Image 59: test

Overall mechanical design is not changed, all parts are in the same spots and same type at the first glance. This is also good news for anyone who have legacy white HP/Agilent/Keysight 3458A DMM. Eventually old instruments do break, and when it’s time for repair and servicing most of the current boards and assemblies can be replaced. Keysight confirmed that all internal assemblies are functionally and physically compatible between old white DMM and new RoHS DMM. I’d expect to have old instruments serviced with existing spare parts stock until that is depleted and eventually everything will be updated with RoHS parts. It is best possible transition between the generations, without scrapping expensive assemblies or whole instruments.


Image 59: test

It is interesting to see that Keysight still kept four additional holes on the steel frame. These are used for special US Navy option to install aluminum shields over the outguard boards, to help instrument pass military EMI/RFI standards. I have acquired set of these shields for some experimentation, but that is the story for another article.


Image 59: Render of two option “SPECIAL” shields for US Navy

Below are photos of how these shields look like on one of my older 3458A units for reference.

p=.
Image 55-55: Top side special shield installed in older 3458A unit


Image 59-55: Bottom side special shield installed in older 3458A unit

These shields are not necessary to meet standard 3458A specifications, hence none of the standard units have them installed. You also cannot buy these shields from the Keysight separately as a part.


Image 59: test

Before going in depth for analog boards, we can take a look on outguard sections.

A6 Outguard power supply assembly

This board connects to A5 digital processor board and A3 ADC board via VersaLink optical interface. It also have all outguard power supply regulators and optical interface for external trigger in/out BNC connectors.


Image 59: test

This PCBA remain unchanged, just updated components for RoHS compliance and minor part changes. It is still using all thru-hole mount stuff, like passives, DIP-chips and capacitors. There is no critical technology needed in this board, so no need to redesign. Microcontroller for interfacing have label with 03458-85512 part number, manufactured on week 19, 2019 and have checksum 25DB. Mains filter is Schaffner FN9222-3-06 dated week 2 of 2019.


Image 59: test

Just like original design, main power regulator ST L296HT is carefully heatsinked to chassis frame with a metal bracket. Little PCB module is small switching DC/DC converter to provide high voltage for vacuum fluorescent display front panel.


Image 59: test

Worth to note massive ferrite block on signal ribbon cable between A6 assembly and front panel assembly. It was installed to catch possible high-frequency noise from high-voltage VFD board and reduce coupling to other circuits around. This is new addition, old HP units did not have this additional ferrite.

On photo above we can also see custom Keysight transformer label again, P/N 9100-4715. It remain unchanged compared to older HP/Agilent 3458A transformers.

A5 Outguard Controller – 03458-66575 (PCB tag 03458-26575, Rev 002)

RoHS-edition Keysight 3458A got brand new controller on new smaller multilayer PCBA. It is running same Motorola (now Freescale) 68000 processor, but this time in small TQFP package, connected to Xilinx ARTIX-7 FPGA. All the glue logic, control logic, memory interface and GPIB IP are now implemented in this rather juicy FPGA.


Image 59: test

NOR Flash chip with E03458-88811 Rev001 label contains instrument’s firmware, also including original Rev.9 firmware that is identical software-wise and operation wise with old instruments. For RAM instrument use three Cypress CY62128EV30LL-45SXI 1Mbit SRAM chips, rated for +3.3V power.

There is 14-pin connector near FPGA, most likely the JTAG debug port for programming and factory testing purposes. This ARTIX-7 FPGA (part number XC7A15T FTG256ABX1913) has 16640 programmable logic cells, 45 DSP slices, 900 Kbits of on-chip memory up to 170 I/O pins and even four 6.6 Gbps transceivers (not used for anything in 3458A). It cost about $30 USD on usual retailers, like Digikey.

Board is loaded with the decoupling caps and has even tiny switching regulator around the FPGA to provide low voltage power and auxiliary rails. According to Texas Instruments part lookup site regulator is TPS62080 1.2A high-efficiency step-down converter in wee 2×2 mm WSON-8 package.


Image 59: test

There is also large lithium coin cell battery to provide power retention for memory, so DMM remembers settings in SRAM when not powered on. Battery connected to Maxim MAX6367PKA31 low-power supervisor designed to provide battery backup for SRAMs. This is a bit concerning, since we see same three SRAM chips powered by limited life battery, similar to very first A5 digital board from 1988. When this lithium cell dies contents of the SRAM would be lost.

Question remains if 3458A RoHS stores all calibration values in one of the SRAMs or in the non-volatile NOR FLASH chip. Since we do not have ownership of this black meter, this theory was not actually tested to avoid possible chances of calibration data corruption/loss. Perhaps Keysight could offer some software utility that can backup and restore calibration data contents in case of battery replacement/power loss. We know for a fact that it is possible to implement such functionality even with original Rev9,2 and old hardware, using undocumented GPIB-commands.


Image 59: test

Only communications interface present in Keysight 3458A RoHS is good old GPIB IEEE-488. Meter can provide maximum datarates up to 100000 samples per second thru GPIB interface, when used in digitize mode and reduced resolution. Again, 3458A does NOT support SCPI protocol, but rather speaks it’s own unique flavour of HP BASIC language.


Image 59: test

It is interesting that Keysight did not move Motorola 68000 processor into FPGA IP-core. Xilinx have D68000 which is software compatible soft-core. I think using true hardware processor was actually faster and cheaper than spending R&D effort to move CPU into FPGA. Also 68000 had to be preserved to guarantee 100% compatibility to original firmware flow and maintain original instrument timings in the code.

Digital board versions (A5 Outguard Controller)

There are multiple revisions of A5 board assembly in 3458A units over its 25 year lifetime.

PCBA Version SKU 03458-66505 Rev.A (66515 Opt.001) 66505 Rev.B (66515) 66505 Rev.C (66515) 03458-66547 or 03458-66548 for Opt.001
First released 1988 ~2000 ~2006
Option variant None and 001 for extended memory
RAM type 2 x DS1235, for low and high bytes SnapHat NVRAMs + SRAM
Calibration RAM type 1 x DS1220, for calibration data SnapHat NVRAM
Solder-free replaceable batteries No Yes

Table 4: A5 PCB versions

A4 Inguard power supply

Inguard power supply board provide isolated clean power to all analog domain assemblies. It remain unchanged, featuring few linear regulators for +18, -18 and +5V rails. There is nothing special on this board, and all components are already RoHS compatible even on older instruments.


Image 59: test

Board output voltage rails connect to A3, A2 and A1 boards. Power output connected via 3M custom ribbon cables to A1, A2 and A3 inguard assemblies.


Image 59: test

A1 DC Board

This is main input analog board, that handle input routing, DC path amplification and corrections, resistance functionality and instrument protection. It had major redesign in terms of used components and layout, replacing most of thru-hole parts from legacy PCB to updated SMT variants.


Image : A1 PCBA overview

For easier signal path understanding I’ve prepared annotated board view with all important circuit blocks. Black arrows show direction of measurement flow.


Image : A1 PCBA overview

Input switching reed relays, A9 module header (and A9 reference itself), main input amplifier and precision shunts still thru-hole components as there is no alternative for these parts in SMT-variant. Overall circuit schematics repeats the old original HP design from 1989, with few deviations where obsolete parts no longer available.


Image : Overall view on A1 PCBA sections

Feel free to click on any of the photos to open high-resolution copy of the image.


Image : Mechanical COTO reed relays for input switching

Relays are custom made for Keysight by COTO, including huge 0490-1657 one for 1kV+ switching. It is not easy to switch very low voltage signals and keep good low-thermal and multiple GΩ leakage performance while keeping protections against 1kV+ possible inputs. For safe switching and long contact life 3458A using interesting scheme to close big high-voltage relay contacts first, then turn on smaller relays. Once signal is measured and level is safe, big relay contacts are open, and low-thermal low-leakage path of small reed relays is preserved. Clever design for high-performance signal path with drawback of the additional control logic to expensive relays.


Image : Input discrete front-end and high-voltage attenuator

Here we can also see high-voltage attenuator divider to provide 100V/1kV ranges, installed near red COTO 3500-0044 low-thermal relay K4. It is manufactured by Vishay, specially for Keysight, part number 1810-1170-1S. This divider have 10 MΩ total resistance, which is what limits input impedance of 100V/1kV ranges. Such design is very common for benchtop DMMs.

Photo on the left shows part of main input DC amplifier, with two hermetic TO-cans and another Vishay film resistor network RP100.


Image : Input amplifier

Amplifier is based around discrete input stage with matched dual Linear Systems LS310 NPN and LS352 PNP pairs. This is final preamplifier before DC signal enters A3 ADC assembly.


Image : Input amplifier, LIS LS31 and LS352 dual transistors

DC signal from other parts of the 3458A input front-end is transferred between sections via shielded coaxial wire and collet-type sockets and PCB pins. This is same design as was used back by HP engineers, who did not trust sending sensitive signal all over with PCB traces.


Image : Reference voltage module and Ohms voltage reference generator

Now to the new changed parts. Pair of old opamp and BB resistor network to generate +10 and -10V from A9 7V reference output is now replaced by SMT-type LT1013 opamp and massive 306E49K50 resistor network, manufactured by well respected Alpha Electronics (part of Vishay Precision Group now). Most likely this resistor network is based around metal foil resistor technology.


Image : Current shunts and internal primary 40 kΩ resistance standard

Another major change is use of the different shunts for all current ranges in new 3458A RoHS meter. Old shunts were pretty good, but nothing exciting to compare with stuff used in other 8½-digit DMMs. Perhaps one unusual aspect is that 3458A have very low current ranges such as 1 µA and even 100 nA. High burden voltages however somewhat limit the low current capabilities of the 3458A in modern world applications.


Image : Close-up on reference 40 kΩ

It is clear that Keysight decided to upgrade original resistors used for all current measurement ranges with improved modern components. Especially peculiar situation, as I have actually thought the same thing good four years before, and did this to my second 3458B unit. That modification was shown and tested in this article from 2016. Not sure if Keysight engineers seen it before, and decided to capitalize on the idea? :)

Here is what that my DIY modification for experimental HP3458A was:

However due to limited testing on older PCB I got mixed results so test outcome was not all that great, as I originally hoped for. Yet another story to continue on another article ;).

Keysight selected different components for the RoHS 3458A:

  • Original R213 replaced with PSN60257 shunt, 0.1 Ω, specifications unknown.
  • Original R212 replaced with 1.5W POWERTRON UNR2-T220 C 1 Ω, max TCR 5 ppm/°C
  • Original R211 replaced with 1.5W POWERTRON UNR2-1410 X 9 Ω, max TCR 5 ppm/°C
  • Original R210 replaced with 1.0W POWERTRON UNR2-1410 X 90 Ω, max TCR 5 ppm/°C
  • Original R209 replaced with VPG BMF S102KT resistor, 634 Ω, max TCR 3.5 ppm/°C
  • Original R208 replaced with VPG BMF S102KT resistor, 4.53 kΩ, max TCR 3.5 ppm/°C
  • Original hermetic VHP101T R207 is the same VPG VHP101T, 40 kΩ, max TCR 0.3 ppm/°C
  • Original R206 replaced with VPG BMF S105KT 500 kΩ, max TCR 3.5 ppm/°C


Image : Current shunts and components around them

Most of these resistors have datecode around 14-16th week of 2019, so we can assume that it took less than a year for Keysight to validate these parts in production design.


Image : High-current 1A current shunt

I would think that Keysight bought enough parts stock to keep production of these instruments in next 5 years, and also to ensure there is enough buffer for quality control and performance metric verifications. But only time and user results will tell us if new parts still good to maintain famous history and stability of original 3458A’s.


Image : Linear Systems LS304 JFET range solid-state switches

Current ranges are switched in by Linear Systems LS304, however it is custom JFET without public datasheet available.


Image : Signal path shielded coaxial jump-wire


Image : Auxiliary circuitry, interesting resistor R15


Image : LS304 JFETs in current source circuitry


Image : Ohm current source, voltage source circuitry and A9 main LTZ1000A-module module


Image : Resistance function overload protection network (black potted hybrid)


Image : Resistance function overload protection, made by Bourns


Image : Current source custom resistor network and JFET switch circuitry


%(imgref)Image : Film capacitors, large SMT resistors for sense terminal protection %


%(imgref)Image : Solid-state ADG211AK switch %


Image : Large MOV for resistance current source protection


Image : Vishay 301358 matched BMF resistors (300 Ω, 3 kΩ, 10 kΩ) for higher current resistance source ranges


Image : Vishay 301358 matched BMF resistors, other side


Image : Precision UXB thru-hole resistors


Image : Control circuitry for resistance current source range switches


Image : LM35DZ temperature sensor. This is what measured by GPIB command TEMP?

A2 AC/TRMS board

A2 comparator changes


Image 12: Keysight 3458A RoHS refreshed 8½-digit multimeter


Image 12: Keysight 3458A RoHS refreshed 8½-digit multimeter


Image 12: Keysight 3458A RoHS refreshed 8½-digit multimeter


Image 12: Keysight 3458A RoHS refreshed 8½-digit multimeter


Image 12: Keysight 3458A RoHS refreshed 8½-digit multimeter

Obsoleted Elantec EL2039 got replaced my little MSOP-packaged Texas Instruments THS4631.

A3 ADC Board

ADC board is what makes 3458A so special. This board was completely redesigned by German Wekomm engineering GmbH. New PCB switched to use of SMT components and replaced obsoleted parts with modern equivalents. Circuit build on 4-layer PCB with lot of ground copper pours to further improve thermal uniformity across the sensitive parts.

Custom ASIC hybrid pin grid array chip U180 is still present in same location. This chip has active silicon die with all carefully designed FET-switches for multislope integration section of the ADC and second die with thin-film resistor networks that are switched in and out of the integrator for high-resolution and high-speed conversions. Thin-film resistor network is most likely tantalum nitride deposition, with laser trimmable sections.


Image : A3 PCBA overview

Enclosing critical circuits into dedicated hybrid help to mitigate possible thermal gradients and limit parasitic to better control. Somewhat similar designs take roots from older instruments like 3456A, 3457A. EEVBlog forum member Noopy recently took beautiful macrographs of ASICs used in old HP 44701A 5½-digit system DMM module of that era. With his permission these images are reproduced below. Click to zoom in for glorious details.

These ASICs have also two separate dies mounted on single sealed carrier. More photos of HP 44701A chips from 1982 and great details shown on Noopy’s site here. Popular 34401A 6½-digit DMM also have two special hybrids, one with resistive network enclosed in purple ceramic and second plastic ASIC with all analog circuitry and switches for multislope ADC.

However 3458A’s U180 Keysight 1NC1-0017 hybrid chip is not used in any other Keysight DMM or instrument, it is unique built to be used only 3458A due to high cost. Old estimates suggest hybrid price around $500 USD a piece. I have restored few old eBay 3458A before, and had to do lot of troubleshooting for ADC operation during those projects some years ago. U180 ASIC hybrid internals shown on Image 44. This photo is taken from old HP 3458A A3 board, but I expect RoHS chip internally to be same design.


Image 4 : U180 internal design. Courtesy bbs.38hot.net forums

Integrity of this U180 ASIC is paramount for 3458A ADC performance. There are multiple cases especially with old HP/Agilent 3458A DMMs from eBay when resistive network inside U180 failed and internal reference ratios are not maintained well anymore. This cause out of spec gain drift and large linearity errors, visible even in short term (hours-days) tests.

Best guess here is that there was bad production run of these hybrid chips with incorrect materials used in assembly. This caused premature failure and excessive stress transfered to dies, resulting issues we seen on faulty units. So far I’ve seen multiple HP-braded 3458A and few Agilent branded 3458A’s instruments that suffered the ADC failure due to U180. Testing for drift and linearity should be a must when buying used 3458A from secondary market, even if it is promised functional by seller. Sadly, only cure for such faulty ADC’s is complete replacement of the assembly, since U180 is quite fragile and impossible to purchase separately.

Hopefully, Keysight had refined the process since and modern manufactured U180 chips got issues limited to minimum. U180 on the RoHS meter was manufactured in 2019, week 3, which aligns with dates on other active components on this A3 PCBA. Either way getting annual warranty service agreement to keep Keysight 3458A’s in check would be a good idea, and save expensive down-time in case of ADC (or any other assembly) failure.

Part of resistive network in U180 hybrid package also used to take external +7.x V from LTZ1000A A9 module and generate internal +12/-12/+5 VDC reference voltages used by integrator switches. External amplifiers for that task are located between U180 and interconnect header J1. These are Linear LT1001 opamps rated for VOS 60µA max, en 0.6 µVp-p and long term offset drift 1.5 µV/month max. Schematics for this part remained same as HP 3458A, other than a shift to SMT type components.

Table outline all versions of ADCs used in 3458A.

Worth to note, not all 3458A’s have same A3 board. There are multiple A3 board revisions, and schematics in CLIP is for one of first revisions. There are some important differences, not reflected in this old schematics.

A3 PCBA Version SKU 03458-66503 Rev.A 03458-66503 Rev.C 03458-66503 Rev.D 03458-66513 New 03458-66513 New-R 03458-66523 RoHS
Board photograph
First released 1988 1989 ~1995 2000 2015 2019
U180 Hybrid Same package and footprint
Interface MCU 8051 Mask 03458-85501 Rev.2 Mask , ATMEL 8051 SMT MCU
U210 Gate array ASIC Fujutsu MB651314 PGA Promex 1820-5770 PGA PCB 03458-26550 with ALTERA CPLD Onboard ALTERA MAX V CPLD
Optical interface HP 1005-0097/HP-1005-0096 Avago HFBR-2521Z/Avago HFBR-1521Z
Integrating cap MLCC 50V 50V black plastic NP0 yellow coated MLCC SMT C0G capacitor
U213 position 74F112PC HP 1820-2924 IC Jumper pin 8 – pin 10 None
ADC U142/U405 comparator EL2018CN DIP8 (HP 1826-1817) Patch PCB 03458-66530 with EL2251CM TI TL3016C + 78L05/79L05 LDO
U404 Ramp opamp Hermetic LF400C Plastic LT1122 SMT LT1122C
Q401,Q402 Hermetic metal can Plastic SMT
C146,C147 capacitors None Present SMT
Additional 10 MHz output None Supported but not populated

Table 10: Different revisions of 3458A ADC PCB Assembly

Part number Description Detail Status
03458-66503 Standard original AtoD PCBA Original board Original, obsoleted now
03458-66513 Newer AtoD PCBA Board with patched comparators/PLD Service only
03458-69503 Refurbished AtoD PCBA Refurbished original board Service only
03458-66523 SMT-version of latest AtoD PCBA, RoHS-compliant Repaired board Current, service only

ADC assemblies are not always available for sale from Keysight service parts store, because the supply is limited and they keep primary stock for instrument repair, which is orderable through our RMA/service process. Price for these modules also vary a lot from $880 to $2k+ USD, depends on grade/type.

Another major change on new A3 related to fast comparators. Original design used pair of Elantec EL2018CN which were prone to failure in many units over the years. These comparators were not easy to replace, because of high ±15V supply voltage and fast switching requirements.


Image 12: Keysight 3458A RoHS refreshed 8½-digit multimeter

Around year 2000 this problem was workarounded by Agilent with Elantec EL2252CM dual comparator. Problem was that this chip sold only in SMT-package. To avoid redesigning whole board, Agilent cooked little patch board 03458-66530 that plug into DIP footprints on original PCB, while functionally equivalent to original pair of EL2018’s.


Image 12: Keysight 3458A RoHS refreshed 8½-digit multimeter

Now all this is replaced by modern Texas Instruments TL3016C ultra-fast 7.6ns low power comparators in SOIC-8 package. To allow their use pair of linear regulators were added right next to the comparator, down-converting +15 and -15 power supply voltages to acceptable ±5V for modern comparator.


Image 12: Keysight 3458A RoHS refreshed 8½-digit multimeter

External sampling clocking on A3

Ability to accept or source reference clock for synchronization purposes is common feature in time-domain analysis instruments. Oscilloscopes, counters, network analyzers, generators can accept external frequency references (often 10 MHz sine or square wave) or provide internal reference clock for external use. For accurate and coherent AC voltage or AC current sampling digitizers can use such reference as well. And because 3458A ADC is well capable to do high-resolution relatively fast sampling this could be helpful in some niche applications. There are external optically-isolated trigger in and trigger out BNC ports, useful for this purpose. But due to nature of ADC control and processing inside 3458A, triggering and sampling have variable delays and true coherent sampling is not trivial.

Perhaps one may think a way to use external clock signal to synchronize DMM’s ADC? It was done before, with special modification to old 3458As to allow sampling with external 10 MHz clock input, but it not trivial and results are mixed. Jitter requirements for such high-resolution ADC are very strict, and many low jitter clock generators are just not good enough in this case. Photo below from CMI NML demonstrate effort required to get close to required specifications. Original 20 MHz TCXO was replaced by PLL HMC1031 + VCTCXO, providing jitter around 1.5 psrms.

%(imgref)Image 1: HP 3458A modified for external 20MHz clocking. Courtesy of Czech Metrology Institute

There is easier alternative. We can use existing 3458A TCXO as a reference signal source, and provide it externally. Then output clock errors and deviations between such modified instruments could be easily measured with commercial counter like Keysight 53230A and samples corrected in post-processing software. So Keysight and Wekomm provisioned this method to have only optically-isolated external clock output for new RoHS 3458A. It is not populated and not available as commercial option, granted that coherent synchronization is still very niche application.

On hardware level we see related new footprints on the PCBA for this additional external 10 MHz clock output from the ADC to allow use of Keysight 3458A for primary metrology lab sampling experiments, such as PTB. There are lot of papers published recently, evaluating AC Voltage metrology and instrumentation around Josephson Arbitrary Waveform Synthesizers. 3458A have flexible and precise AC digitize functionality. Thus external clock output is helpful to allow true coherent frequency-locked sampling.

Some of the examples of such work published here and here. Keysight 3458A synchronization needs are also used for power quality standards and power calibrator research. EMPIR Joint Research Project 15RPT04 TracePQM is one of such efforts using Keysight 3458A DMMs in this field.

Because ADC and DMM front end is floating it was not possible to provide simple galvanic coaxial connector for such an output. Optical transmitter is added to resolve this caveat, same type fast optical VersaLink interface between ADC and main processor.

These parts are unpopulated on production board, but it is easy to guess the parts by looking at very similar circuit for Avago HFBR-1521Z optical transmitter. Modification would involve parts from Table. Perhaps instead of 5 Mbit HFBR-1521Z one need to use faster transmitter, because output clock from ADC would be 10 MHz.

Reference designator on PCB Part number Digikey link
U10, optical transmitter Avago HFBR-1528Z Broadcomm HFBR-1528Z
CR10, diode 1N4150W Vishay 1N4150W
R5, resistor 150 Ω , 1206
U8, logic driver/inverter 74HCT14, SOIC14 ON MC74HCT14ADR2G
L1, ferrite bead for power 0603 inductor TDK MLF1608DR33MTA00

A9 DC reference PCBA

This small module is the key assembly for long-term stability of the multimeter and calibration longevity. It is designed around best solid-state reference zener money can buy, Linear (now part of Analog Devices) LTZ1000A. Circuit repeats slightly modified reference schematics with LT1013 operational amplifier and few VPG BMF precision high-stability resistors. LTZ1000A Ultra-Zener is covered with plastic cap from both PCB sides, to prevent airflow and thermal gradient changes. This reference design provide unbuffered 7.x VDC output, with typical noise below 1.5 µVp-p and have excellent temperature stability <0.05 ppm/°C.

There are multiple grades of selected 3458A A9 PCBA, listed in Table. All grades have same components and parts, the difference is only in treatment and selection process. Electronic components even in the same bath often have slightly different stability. Keysight tests each reference and assign a grade based on the achieved performance parameters for annual drift spec.

Part number Description Stability spec Status
03458-66509 DC Voltage reference, standard <8 ppm per year Current, $699 USD
03458-66519 DC Voltage reference, Option 002 <4 ppm per year Obsolete, $997 USD
03458-66529 DC Voltage reference, HFL <2 ppm per year Obsoleted in 2019, was $2k USD
03458-66539 DC Voltage reference, standard, RoHS <8 ppm per year Current, $533 USD
03458-66549 DC Voltage reference, Option 002, RoHS <4 ppm per year Current, $533 USD
03458-66559 DC Voltage reference, LF, RoHS <8 ppm per year Current, $585 USD
03458-80003 DC Voltage reference, Option 002 <4 ppm per year $997 USD
03458-80009 DC Voltage reference, STD <8 ppm per year $669 USD

DMM we had for test have standard RoHS reference, part number 03458-66539. I am not sure what is LF designator in 03458-66559 means.

LTZ1000A die is still being ran at high temperature, close to +90 °C in all of these references, including RoHS version. This is a sacrifice choice for wide operation temperature range specified for both old and new 3458A DMM. As result of high temperature annual drift is degraded a lot, also reflected with standard specification of 8 ppm per year.

In real world usually stability is better, especially for well cared instruments ran in stable lab environments, with frequent fan filter cleaning, etc. Based on multiple test results of older 3458A owners it is not uncommon to see standard (non-002) DMM to have less than 4 ppm per year drift.

However it is possible to improve that with simple modification, reducing LTZ1000A die temperature to ~50°C. This modification is shown here in detail. My two special 3458As rebuilt from of standard units both have modified A9 reference modules and show annual drift less than 2 ppm/year.

Using Keysight 3458A RoHS to boost capabilities of multi-function calibrators.

Features wish-list for future Keysight Metrology level DMM

There is always a desire for more, and precision long-scale multimeters are not exception. We have used 3458A and other instruments of this league for many years, so perhaps some of the improvements could be suggested for the future new metrology DMM. It’s easy to say “we want everything better” but could be a challenge to actually highlight what is key importance and what is not very critical. Of course each target application has own demands, but for us wish list is summarized below:

  • Primary DC reference treated for better long-term stability. High-temperature set point (as by-product to +50 °C ambient spec) for LTZ1000A A9 PCBA (on both original and RoHS DMM) is the main reason of poor 8(4) ppm/year DCV performance specification. Following conventional design with oven temperature around +60°C would yield 2 ppm/year without significant redesign.
  • Maintain ultra-high linearity. Good samples of the existing 3458A can already achieve better than ±0.05 ppm INL on base DCV range. Only wish here is to keep and preserve this vital specification in new design. This is tough challenge, but important one and one of the reasons why 3458A is still industry standard in field, 32 years since it’s launch. This is also vital to Artifact Calibration performance.
  • Electrical front/rear switching. Many applications where long-scale DMM is needed require ratiometric measurements. Programmable input terminal selection can help all these cases. Old Fluke 8508A/Datron 1281 or even Advantest R6581T can do this, so it’s not unreasonable feature to expect.
  • Scanner expansion port. Keithley managed to cram whole expansion card port into half-sized 8½-digit Model 2002. With much bigger full-width 19” instrument, smaller size PCBs and modern SMT parts it should be possible to add expansion card to next DMM project. Guarding could be tricky, but can be resolved by optional enclosure.
  • 200% overrange for all functions. Measurement of signals that a slightly over 120% range without performance penalty. Currently 3458A ADC input scale is limited to ±12V full scale, however with some smart ranging this probably could be workarounded. Competitor meters already provide 200% ranges.

Summary and conclusion

Based on the work done here we can confirm that legacy of legendary HP 3458A 8½-digit DMM still preserved and lives on by latest RoHS compliant version refreshed by Keysight. It allows continued support and use of vast infrastructure and countless of systems with integrated HP 3458A. Performance and remote control command set that 3458A have are rather unique, so it was one of important factors considered in refresh project.

I would like to thank Keysight Technologies team for providing production DMM demo unit for this testing and making this little project possible. It does matter a lot for community to see manufacturer commitment to such special case like independent testing with expensive product. I would like to thank Rado Lapuh, fellow metrologist and author of excellent book about Keysight 3458A sampling and AC capabilities. Reading his book inspired me to look closer into our instruments and start a new exciting project for AC metrology.

This work and results also would not be possible without constant help from EEVBlog metrology section community. Special credits to go fellow USA xDevs.com Lab maintainer Todd M. and his decade-long efforts to keep the hobby-type lab in check for absolute accuracy. I have used his standards to perform transfers for Volt/Ohm during August 2016 meetup and we cooperate on daily basis to do inter-lab comparisons of our equipment performance.

We wish to see more of such work done also at hobby levels among engineers. Publications, even about less precise bits of equipment often help students and engineers to share and preserve complex knowledge about analog electronics and advance science of measurement in general. Sometimes that include repair or use of 30-year-old aged but still very capable secondary market instruments and references.

Until then, stay tuned and let us know your feedback! If you have any suggestions or corrections to make, feel free to reach out to us. Discussion about this article and related stuff is welcome in the comment section or at our own IRC chat server: irc.xdevs.com (standard port 6667, channel: #xDevs.com). Web-interface for access mirrored on this page.

Projects like this are born from passion and a desire to share how things work. Education is the foundation of a healthy society - especially important in today's volatile world. xDevs began as a personal project notepad in Kherson, Ukraine back in 2008 and has grown with support of passionate readers just like you. There are no (and never will be) any ads, sponsors or shareholders behind xDevs.com, just a commitment to inspire and help learning. If you are in a position to help others like us, please consider supporting xDevs.com’s home-country Ukraine in its defense of freedom to speak, freedom to live in peace and freedom to choose their way. You can use official site to support Ukraine – United24 or Help99. Every cent counts.

Author: Ilya Tsemenko
 Created: Dec. 13, 2019, 4:05 a.m.
 Modified: May 21, 2024, 9:19 p.m.

References

  1. Rado Lapuh : Sampling with 3458A", ISBN 978-961-94476-0-4
  2. Keysight : 3458A Product page
  3. EEVBlog : 3458A Black Edition
  4. Traceability Routes for Electrical Power Quality Measurement site
  5. QWTB - Software Toolbox for Sampling Measurements
  6. TPQA-Traceable Power & Power Quality Analyzer
  7. Martson, David. (2002). The effectiveness of artifact calibration in computing internal resistance values.