Retrofit DMM HP 3457A

[MiDi]

I got a HP 3457A in good condition for reasonable price at ebay.
It performs nicely, but I thought it would be a good investment to retrofit the 30 year old electrolytic caps for power supply rails, the mains input filter and the battery for calsettings.

For others who want to retrofit this equipment too, here is what I have done:

I turned it apart to see in what condition the unit was and I was really surprised how good it was.
No dust, no visible aging and the best: battery changed in 2009 
No battery change needed for the next 10-20 years - nice!
Battery replacement would be really critical, as it holds the calsettings and the replacement method is not so easy...

It turned out, that there are only four caps inside with standard sizes, so no worry about:
3300µF 7,5mm (digital board)
1000µF 7,5mm (analog board)
2x330µF 5mm (analog board)

I replaced them with Panasonic FR series caps (105°C 10.000h lifetime), this should last for the next 30 years - even at higher temperatures 
The sizes are not critical, because today you get the same or even better ratings at half the size, so only the lead-wire spacing has to match.
It would be possible to replace with higher capacitance at same or smaller sizes.
The old caps are still in tolerance, but as lacking of an ESR-Meter, no further investigation is possible.

The mains input filter from Schaffner has bad reputation as there are multiple reports of blowing or burning units and as there is no fuse in front of it, this is really a critical component.
I replaced it with a Schurter 6A type (Schaffner has 3A) with matching component values, but its a bit longer in size.
It turned out that there are no problems with cable length or other components with this size, but it could be a little problem if you leave the 3300µF cap...
X2: 47nF
Y2: 2.2 nF
L: 0,8mH (Schaffner 0,75mH)

Now the unit is refreshed and reliability is improved.
The HP 3457A is very nice featured for its age - I like it! 

What do you think:
What could be other reasonable retrofit for getting better reliability?
And for getting better performance?

[alm]

Apart from the mains input filter I am not aware of any major reliability issues. In a well designed instrument, no single component upgrade should give you a major upgrade in accuracy (i.e. many components contribute about evenly). If you were TiN, you might try to upgrade the LMx99 voltage reference to an LTZ1000-based solution .

A possible modification I would consider is to add a backlight to make that horrible LCD a bit more readable.

[MiDi]

You are right, the old caps are reasonable good parts and sealed with epoxy at the bottom - never seen this before.
But if they fail, there would be a major impact of electrolytic in the surrounding - a nightmare to clean up.
Even if they run out of specs without blowing and degrading your supply rails over time could lead to inaccurate measurements you would not recognize instantly.
New caps cost only a couple of bucks - even the best ones, so it is worth to replace them and do not care anymore.

Theoreticaly all components should be matched for overall performance spec, but in reality I do not think that this is the case.
Cost and available technology could lead to parts that dominate the whole performance spec.
I do not think that 30 years ago they had opamps with specs like LTC2057 (introduced 2013!) for affordable cost - to give an example.

I have read the LTZ1000 and LM399 threads and am really impressed of all the volt nut voodoo stuff.
But a LTZ1000 in this kind of DMMs is not worth it - it doubles the value, but it does not lead to a magnitude better in specs.
For that performance get a metrology grade DMM 

There are two options for upgrade to a lit display, EL-foil and edge LED.
EL-foil seems to lack for brightness, edge LED has a high risk of damaging the LCD.
If the LCDs would be available for a couple of bucks it could worth a try...

[alm]

I am not sure modern components would have made much of a difference. Its predecessor, the HP 3456A, introduced in 1981, had a 10 V DC 1 year accuracy of 23 ppm after some small revisions (including switching the reference to an LMx99 from a custom HP one). The HP 3457A, introduced in 1986, had a 3 V DC 1 year accuracy of 25 ppm. The Keysight 34465A, introduced in 2015, has a 10 V DC 1 year accuracy of 30 ppm. Not a lot of progress in ~35 years, is it? It is mostly the digital side (and maybe ACV?) that is improving.

I am not convinced that upgrading the reference will give that much of a benefit either. But it is interesting that all three meters use the same reference, and that the specs of the 3456A were tightened substantially after more data was available on the behavior of its voltage reference.

[Kleinstein]

The 3457 meter has a kind of odd construction that there is no generic DCV range that fits well for a 7 or 10 V reference. The best range is 3 V and the 30 V range already uses an input divider. So it is still kind of made for a 1.1xx V standard cell and not a modern 7 V zener reference.

The critical amplifiers use JFET pairs - kind of high expensive construction by today's standard, but performance of this is still good. There might be cheaper and simpler alternatives with monolithic JFET OPs today with only slightly lower performance. So I don't see sensible upgrade options (if at all it might be with the ohms ranges where they use just OPs). A big design problem of this meter is the odd construction of the input amplifier that does not offer a gain of 1 and is thus missing the possibly best range. The ADC design also has some possible limitations for the INL. But these two point are nothing to change afterwards.

So it is a reasonable good instrument, but not a great design. The nice thing is more like the option for an internal scanner card. Overall the meter is more like noisy - so maybe on par with a Keithley 2000 or 2001 (which is also kind of noisy).

[MiDi]

I am not sure modern components would have made much of a difference. Its predecessor, the HP 3456A, introduced in 1981, had a 10 V DC 1 year accuracy of 23 ppm after some small revisions (including switching the reference to an LMx99 from a custom HP one). The HP 3457A, introduced in 1986, had a 3 V DC 1 year accuracy of 25 ppm. The Keysight 34465A, introduced in 2015, has a 10 V DC 1 year accuracy of 30 ppm. Not a lot of progress in ~35 years, is it? It is mostly the digital side (and maybe ACV?) that is improving.

I am not convinced that upgrading the reference will give that much of a benefit either. But it is interesting that all three meters use the same reference, and that the specs of the 3456A were tightened substantially after more data was available on the behavior of its voltage reference.

Seems to be a well balanced design, as there is very low benefit from modifying anything as claimed in the article on xDevs 

[Kleinstein]

I do not consider the 3457 a good design. It is more like an odd one with the poor choice of not having a range suitable to directly measure the 7 V reference.  As the voltage range is set by the input hybrid there is no easy tweak. So it kind of a surprises it still works that well despite of the poor general design at the input. The meter could have been quite a bit better with an input stage / switching made for 10 V primary range.
I would consider the 3446x specs a lot more conservative than the 3457 specs, especially for longer time.

The 3457 is not a direct upgrade from the 3456. It is more a line 3455 - 3456 - 3458  made for metrological use. The 3457, 34401, 3446x are more for more general use. For higher accuracy general use there is the 34470 now.

The later 3456, 3457 and 3446x and many other 6 digit DMMs still use the same type of LM399 reference. So no change there, though modern DMMs  could profit from something in between the LM399 and LTZ1000 (e.g. a LM399 variation that runs with higher zener current and slightly lower temperature - half the drift and noise would be a nice target).

[wiesl]

Which HP multimeters do you recommend?

[MiDi]

HPAK3458A
Depending what you want to do with it, there are cheaper ones