PX Reference

[cellularmitosis]

Current status:

Current board: v2.4.1

OSHPark: https://oshpark.com/shared_projects/hUZKe65r

Github: https://github.com/pepaslabs/px-ref/tree/master/kicad/releases/v2.4.1



Original post:

It lives!

Off to a bit of a rough start, though:

• Tempco appears to be roughly 5uV/C (nearly 1ppm/C).  Yuck .  I haven't populated the 400k resistor yet.
• While probing around the board I accidentally applied 15V to +Vz and fried one of my LTZ's.  Oopsie.   
• I initially soldered in the R4/R5 divider backwards.  Sheesh.   

I decided to socket the first board so that I could test each of the LTZ's.  It looks like this first one might be jumping by a couple of microvolts.

The critical parts were all from the ebay seller hifi-szjxic, so it was nice to confirm that he sells working parts.

• U1: LTZ1000A, date code 1237
• R1: 120R RNC90Z (2ppm/C)
• R2, R3: 69K8 RNC90Y (5ppm/C)
• R4/R5: 12K2/1K VSR144T (2ppm/C)

The other resistors are cheap metal film "1%" which I had on hand.  They might not be 100ppm/C as HP specs for the A9 board.

I made a mistake on the board and left out Andreas' C15, so I had to bodge that in on the underside of the board.

A couple of the caps are ceramic, as I didn't have 10nF and 22nF in film on hand.

When I cleaned the board with some IPA it seems to have left a slight white film.

Overall, very glad to finally be making some real steps here!   

Thanks so much to the members of the forum!  I have learned so much already!

https://github.com/pepaslabs/px-ref/tree/master/kicad/releases/v2

edit: oops, intially attached the wrong graphs.

[hwj-d]

Hi cellular,
thanks for your very interesting postings and work. 

Which PS do you use? After some burn in my kx, I'm using pure new unsecured 4s 18650 that gives me a 4 day continuously testing time down to 12v before i change them to a fresh charged set. I see a comparatively cleaner VRef as with my DP832.

[cellularmitosis]

Interesting!  Right now I’m just using a cheap linear bench PSU. I had an idea of buying a small 12v lead acid battery to get a day or two of runtime within a shielded enclosure.  Sounds like you are doing the same!

[Andreas]

Tempco appears to be roughly 5uV/C (nearly 1ppm/C).  Yuck .  I haven't populated the 400k resistor yet.[/li][/list]

I would first remove the socket before I populate a resistor.

With best regards

Andreas

[SilverSolder]

I know what you mean about IPA leaving a white film behind.   You can buy "clean" IPA (e.g. from MG Chemicals) that is of a better/cleaner grade.

[TiN]

I don't know what socket is used here, but on my 2 boards with AUGAT sockets (pin contacts are colet type, same as high-quality DIP panels) there is no visible difference in tempco, due to socket.

Compensation resistor can help to improve tempco from 0.1-0.2ish ppm/K to <0.05 ppm/K. If you see 1ppm/K, something is very much off, so need to fix that first.

in the R4/R5 divider backwards.

Jeez, If that's the LTZ which is in board now, that likely caused problem for LTZ too. It takes seconds for A-chip to reach setpoint temperature due to 400 C/W thermal insulation between little die and metal can.

PSU need to have high isolation from mains and avoid ground loops. It's easy to check if you have power issues: connect LTZ output to meter in normal polarity, record reading. Then reverse output connetion to connect LTZ output in reverse polarity. Absolute voltage level should not change. If it changes, you have power problem or parasitic ground loop that steals current.

[cellularmitosis]

Luckily I caught the R4/R5 mistake while I was still assembling the board

[cellularmitosis]

v2.2 of the board.

Changes:

• Smaller capacitor footprints, tighter layout
• Added an output buffer (LTC2057).  Thanks to Andreas for inspiration, see https://www.eevblog.com/forum/metrology/cx-reference/msg1381896/#msg1381896
• Added a jumper to optionally tie ground to the enclosure via brass standoff

OSHPark: https://oshpark.com/shared_projects/etqxI0Ok see follow-up post below.

github: https://github.com/pepaslabs/px-ref/tree/master/kicad/releases/v2.2

edit: attached schematic

[Andreas]

Hello,

there is a "bug" in the schematic:
R31 has to be connected to the other side of R32. (directly to the output connector).
(like R22 in my schematic above).

with best regards

Andreas

[cellularmitosis]

Andreas, thank you so much for taking a look. 

If I could bother you a bit more, could you explain why?  I was a bit puzzled about the placement of that resistor as well as your 10nF capacitor.

[kj7e]

Andreas, thank you so much for taking a look. 

If I could bother you a bit more, could you explain why?  I was a bit puzzled about the placement of that resistor as well as your 10nF capacitor.

Looks like you may have implemented part of Andreas 10v gain in to your unity gain buffer.  R31 and C31 is not needed for unity gain, only needed if you are going to also use ~25K resistor to ground between R31 and pin 2.  This would give you 7 to 10v gain.   It looks like you just want 1:1 here though.  So just tie pin 6 to pin 2.    R32 is just an inrush current limiter for C32.

[Andreas]

If I could bother you a bit more, could you explain why? 

see e.g. here (Figure 17)
http://www.ti.com/lit/ug/tidu032c/tidu032c.pdf

in your case with a 10 Meg Load you will have 2 ppm DC error because the series resistor which isolates the capacitive load is not within the feedback path.

R31 and C31 is not needed for unity gain,
 So just tie pin 6 to pin 2.   
 R32 is just an inrush current limiter for C32.

no
no
and no

please read the application notes from the Op-Amp manufacturers.
"isolating (large) capacitive loads"

with best regards

Andreas

[cellularmitosis]

Thanks Andreas!

kj7e, this circuit is essentially a combination of an input bias compensation technique
 (equal resistors on the inputs of a buffer, see https://www.allaboutcircuits.com/textbook/semiconductors/chpt-8/op-amp-practical-considerations/) and the capacitive load technique Andreas linked to.

I have also read that some op amps need resistors in the buffer configuration for stability reasons (e.g. the AD797).

[cellularmitosis]

Updated v2.2 board with Andreas buffer corrections.  Also, I added a small copper zone around the LTZ pins to encourage constant temperature in this area.

OSHPark: https://oshpark.com/shared_projects/5n56VbPg

github: https://github.com/pepaslabs/px-ref/tree/master/kicad/releases/v2.2

edit: corrected out-of-date bottom-copper.png

[TiN]

I'd not leave those floating capacitors around LTZ .
Also what is the reason of thinny-skinny neck to the "star" points?

[cellularmitosis]

I'd not leave those floating capacitors around LTZ .
Also what is the reason of thinny-skinny neck to the "star" points?

C11 and C12?  Are you concerned they will disrupt the circuit, or are you concerned about their placement (that they will get hot)?

Initially I had the star traces full-width, but when they diverge at small angles, this creates a large area of overlap, and I was concerned that this was not really a single-point ground and also that it would be difficult to solder.  But I don’t have a lot of experience with star grounds, so maybe this is I’ll advised.

Edit: “hot”, not “hit”

[cellularmitosis]

In order to maintain stability, the RC time constant of R_f*C_f must be (at least) twice that of the RC time constant of R_iso*C_load.  This is easy to simulate on LTspice.

Ah, interesting.  10,000R * 0.00000001F is much larger than 22R * 0.0000001F, so I should be stable.

[TiN]

No, I meant those unconnected PCB shapes. And on 2 layer PCB everything solder easily  .

[cellularmitosis]

Getting some results which are a bit better now.

I've put the reference in a Hammond 1590B aluminum case, used some "cotton rounds" below and a cotton ball above the board to protect from air currents (this might be too much insulation, we'll see!), and I've connected the case to the negative terminal of the power supply.

For comparison, I also put an LM399 board in a Hammond case.  The LTZ is a bit less noisy, but not as good as the datasheet spec of 1.2uV p-p.  However, my 34401A uses an LM399 reference, so I may have reached my measurement limit.  Next step will be to populate a second board and compare them back-to-back on the 100mV range.

[Dr. Frank]

For comparison, I also put an LM399 board in a Hammond case. 
The LTZ is a bit less noisy, but not as good as the datasheet spec of 1.2uV p-p.  However, my 34401A uses an LM399 reference, so I may have reached my measurement limit. 

You can hardly judge the LTZ1000 noise by using the 34401A.
Instead, you definitely measure the noise of the LM399 inside the instrument.

[Kleinstein]

When measuring the external reference you get a combination of the noise if the external and meter internal reference. At higher frequencies there might be some filtering, but usually not much. In addition there will be noise from the meter (e.g. ADC) itself. To a certain degree on can subtracts the meters own noise, but this is limited.

For the reference in the case it might be a good idea to have some filtering for the voltage input, something like a cap directly across the terminals and a ferrite ring as a common mode choke.

[Andreas]

(this might be too much insulation, we'll see!)

without a DEWAR? Never!
most of the heat is transported by the pins and the PCB.

with best regards

Andreas

[cellularmitosis]

OK!  Got a second board assembled and have set them up back-to-back.

I'll be out of town next week, so I'll just let it run the whole time (and ssh in remotely to check up on it  ).

Also, Dr. Frank's boards just arrived.  Too bad Rhopoint hasn't gotten back to me about those econistors yet!

[cellularmitosis]

This is by far the quietest graph I've ever recorded

[Andreas]

OK!  Got a second board assembled and have set them up back-to-back.

you should use 2 galvanic isolated power supplies for that.
Otherwise the ground currents (heater) influence the voltage measurement.

With best regards

Andreas