MX Reference

[ManateeMafia]

Once in a far away galaxy a volt-nut was born. As typical, collection of useful and not, expensive and cheap, rare and common test equipment gear and accessory grew, as the volt-nut explored the world of fleabay. But no volt-nut journey will be complete before the final unique artifact is built. Being such, allmighty Repus-Renez reference, which is beginning and end of all the references!

..oh, it's EEVBlog here, people here don't like watery intros.. straight to the meat now.

Project goal and specifications

* Maximum stability for DC voltage reference.
* Low tempco precision wire-wound resistors, PCB to accept for both PWW Ultrohm Plus and Vishay VPG resistors.
* Linear LTZ1000 ultrazener URL.
* Thermal sensor NTC for external calibration purpose.
* Compact 80 x 120 mm form factor to fix Hammond metal enclosure.
* Thermostat setpoint +55C.
* Kelvin sense output.
* Single-rail voltage input.
* PCB space for add-on board to use direct output from LTZ1000 for further down/up-conversion.
* Basic reverse supply voltage protection.
* Room for 3D printed cap on top and bottom of board for LTZ
* Room for 3D printed cover for rest of board.

Typical schematics from datasheet, which is a proven baseline with already great performance. This datasheet schematic well capable to surpass any precision DMM in long-term stability. Hence it's used by many proven instruments, such as HP 3458A, 34470A, Keithley 2002 with just minor tweaks.



Now to stay on the safe side on this project, base reference circuit will repeat datasheet schematics almost exactly. No controversial chopper amplifiers, no voodoo magical stress-relief slots, no exotic PCB materials. This will still have good learning curve, for one who only took apart instruments with these references, not built from scratch.


   
The schematic is straight-forward, and was explained and covered on EEVBlog many many times. Resistors of choice this time are custom PWW's or more expensive VPG BMF's. Opamp is recommended LT1013, which performance is well matched to design goals of LTZ module. Most of noise and drift on the output voltage will be due to own LTZ1000 instability, than used opamp.

Diode D3 provides reverse polarity protection, as no matter how careful you are, one day you will connected positive PSU output to the ground reference of the module. "Woops" would be rather sad portmortem note for the 55$ USD zener.

Layout design

Layout is done on two layer FR4 PCB with "Dangerous Red" solder mask color, as should be for the prototype. To give it a nice touch, exposed metal pads have electropated ENIG coating. This will let one to assemble board without soldering issues even after a year of storage.



Almost all components are thru-hole, which makes assembly a breeze, compared to much smaller and difficult to assemble xDevs.com KX module. This was fourth or fifth iteration of the layout, since this was my near entry to electronics design and manual layout. First iterations were not worthy of even the "Drunken spider" nickname, so it's best we don't see them here...



This PCB designed specifically for Hammond 1457K1202 metal enclosure, which is large enough to feature future upgrades with 10V scaling circuitry and other possible additions to design. It's important to select good form factor with some room to scale up, than scratching head how to jam in that extra capacitor needed.

Assembly notes

Other than traditionally long lead times for precision custom resistors, assembly and rest of components are straight forward. Total BOM list of components used with Digikey P/N is listed in table below.

Digikey Part Number	Manufacturer	Manufacturer Part Number	Customer Ref Number	Customer Description	Quantity for module
N/A	Ultrohm Plus	N/A	R1-R5	Custom made wirewound set	1
495-2479-1-ND	EPCOS (TDK)	B32529C1104J289	C1,C2,C4	0.1uF Cap Film 5% 100VDC	3
495-2489-1-ND	EPCOS (TDK)	B32529C1223K289	C3,C6,C7	22nF Cap Film 5% 100VDC	3
399-3905-1-ND	KEMET	T495X476K035ATE300	C5,C8	47uF Cap Tant. 10% 35V	2
1N4148FS-ND	Fairchild/ON Semiconductor	1N4148	D1-D3	1N4148 Gen Purp DO35	3
PZT3904CT-ND	Fairchild/ON Semiconductor	PZT3904	Q1	Transistor NPN	1
1.00KXBK-ND	Yageo	MFR-25FBF52-1K	R6	1k ohm 1/4w 1%	1
10.0KXBK-ND	Yageo	MFR-25FBF52-10K	R7	10k 1/4w 1%	1
1.00MXBK-ND	Yageo	MFR-25FBF52-1M	R8	1M	1
402KXBK-ND	Yageo	MFR-25FBF52-402K	R9	402k ohm 1/4w 1%	1
10.0XBK-ND	Yageo	MFR-25FBF52-10R	R10	10 ohm 1/4w 1%	1
BC2299-ND	Vishay BC Components	NTCLE203E3103FB0	R11	Thermistor 10k	1
LTZ1000CH#PBF-ND	Linear Technology	LTZ1000CH#PBF	U1	Shunt Voltage Reference	1
LT1013ACN8#PBF-ND	Linear Technology	LT1013ACN8#PBF	U2	IC OpAmp	1
HM1023-ND	Hammond Manufacturing	1457K1202BK	N/A	Enclosure, Black	1
HM1022-ND	Hammond Manufacturing	1457K1202	N/A	Enclosure, Natural	1
1457K1202EBK-ND	Hammond Manufacturing	1457K1202EBK	N/A	Enclosure, Black EMI Shielded	1
1457K1202E-ND	Hammond Manufacturing	1457K1202E	N/A	Enclosure, Natural EMI Shielded	1

Linear parts were purchased directly from their website shop, custom resistors ordered directly, and rest of parts come from Digikey USA. The initial build was made using Riedon USR series resistors that were readily available. Several sets are on order from Edwin and a board will be assembled when they arrive.

Assembly passive components, such as capacitors, diodes, resistors first. LTZ, opamp and precision resistors install last after board is cleaned and checked, keeping soldering times minimal and keeping parts cool to reduce thermal stress.











MX Voltage reference module designed for education and hobby experimentation purposes only, which resulted in used components choices. If one need ultra-stable voltage reference for production device or practical equipment, very different approach would be required. This includes adding necessary polarity/overload/ESD protection circuits to the input and outputs of the reference, etc.

To reduce thermal stress to foil resistors, heatsink legs with small pliers or braid during soldering, so resistor body would not heat up as much. Excessive heat on resistive element will cause stress and possible hysteresis, which will require long-term recovery to original state afterwards. Make your soldering brief and short. If solder joint not good, wait till board and resistor cool down back to ambient and then try brief reflow for bad connection again. Do not keep iron tip heating up resistor lead longer than few seconds.

Initial data results

Connect power to the modules from quiet (electric noise wise) linear power supply, leave to settle for few days/weeks/months and then measure with best stability multimeter available.

Here are some measurements and test data completed on LTZ1000 module, with alternative resistors, as lead time exceeded amount of patience I had:

Test result description

Test setup

Datalog

 
  Data MX1, initial       3458A NPLC100, PD PSU +15V  MX1 Test 1 realtime log 


Or in graph form:



If you interested in this project build or have any questions, feel free to jump in comments. We also have more detailed and less traditional "KX LTZ1000-reference build":https://xdevs.com/article/kx-ref/ so be sure to check that one out as well.

[cellularmitosis]

Fantastic write-up!  Always great to see a published result, and more participation in this pursuit.   

[ManateeMafia]

Thanks, there was a lot of "influence" in this article from a fellow member.   

The first rev has brought some ideas and changes to this layout and it is waiting based on test results. The free space on the next board will almost double.
I will be building a thermal chamber to house the board and run it through some testing similar to how others have tested their designs.

[Andreas]

Hello,

just some annotations:

C7 would work much better if you place a series resistor (10K) between R1 and C7.

Do you plan to use the 805 style (6*13mm) of Edwins resistors or the smaller type that I have seen in other (KX-based) designs?

For the 805 style (as well as UPW50 or 8G16)  the footprint of the resistors should be minimum 700 mils for the hole distance.
For me it looks that you have only 600 mils.

With best regards

Andreas

[ManateeMafia]

Good catch on the measurements. I realized the size error when the boards arrived. I am certain I used the measurements of the smaller resistors in the TC Measurements thread when creating the footprints. My library has been updated and the Rev A01 board layout has been shifted to accomodate the spacing.

For now, I have a second set of BMF resistors for another board. I would hate to bend any PWW resistors into an unforgiving shape.

[MK]

If you place a BF245C at the output of the LT1013 then the diode d1 is not needed, as it becomes self starting, and as a bonus any extra current drawn does not heat up the lt1013, so that stays cooler so the offset does not change as much, and a current limit is built in.
just a suggestion for future versions

[quarks]

Thanks a lot for sharing

[TiN]

Since I was trusted with hoarding full data, here's plot from the beginning:



It's too early to tell if there is problem with the build or not, need to log for another few weeks to be sure, if we reach the stability.
Just a reminder for newcomers that ppm-level measurement do NEED weeks of undisturbed test time, accuracy in setup, and above all - patience.

There is ACAL +/-0.3C for 3458A, to ensure measurement stability to own meter's LTZ level.
Orange line is BME280 temperature.

RAW DSV-file for the analytic minds.

Green curve is bonus, stability of Power Designs 2005A ovenized power supply. It set to output +15V to power MX module, and drifted up only +42 ppm since the beginning. Output noise is only few times worse than LTZ output itself, so I'm quite impressed with this vintage gem (expecting paybay prices to rise now!).

Another idea for the cal-clubs - have two level loops of rovers. Basic level loop with usual LM399/non-ovenized zener references etc for quick turnaround and comparisons for <6.5d meters and slow "Volt-nut grade" loop with LTZ1000/LTFLU references with 1 month/1 member turnaround time for 7.5+d meters .

[Kleinstein]

The 120 Ohms resistor has quite a lot of long thin line in series. My feeling it the copper resistance might get relevant. There is also quite a large loop area - so unless in a good case, it might tend to pick up radio signals.

The JFET at the output of the OP, does not really offer an extra current limit. With a positive gate voltage the current can go higher then IDSS and the gate current (from the OP) can also add to this. So for a current limit it would need a resistor in the drain line and maybe at the gate.

[Andreas]

or perhaps better only one resistor at the source like here?
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/?action=dlattach;attach=52101

With best regards

Andreas

[ManateeMafia]

@Andreas

Thanks. Are you referring to R13?

[Andreas]

no I refer to R18 (the 220 R)
R13 (47K) is only a "startup" resistor. (but of course optional when the BF245 is populated).

By the way the BF245 is no longer manufactured.
So you may want to use the SMD-version (SOT-23) BF545C instead.

With best regards

Andreas

[ManateeMafia]

I discovered that the BF245C was discontinued last night after MK recommended it. I will add the modifications to rev A01.

[dr.diesel]

Bump, any updates?   

I've got some spare Edwin resistors itchin for a purpose.

[TiN]

Bump-bump, can't wait for tempco results [emoji14]

[ManateeMafia]

It has been burning in since it was built. Data collection has been problematic with a bad SD card on the RPi but the board has shown signs of settling.

I have some sets of resistors coming in from Edwin eventually so I hope to get the next rev made and tested before they arrive. I still have to make a few more changes before I send out the gerbers.
The next build will be with the LTZ1000A.

I have all the parts for the test chamber so I need to go ahead and assemble it.

[RandallMcRee]

Very nice!
If this is a kicad project would it be possible to post the kicad project folder? Would help me (and presumably others) to jump-start our efforts.

Thanks,
Randy

[ManateeMafia]

Thanks.
This one is in Eagle CAD. One day I will make the jump but I don't have a need for it yet.

Once I get the next rev tested, I won't mind sharing what I have. I am certain there will changes based on what others have recommended and already done.

[TiN]

I think she's ready for the tempco testing.

Attached plot of MX1 prototype stability over 1015 hours since power on.

We can see rather harsh voltage drop first few hundred hours. Last 100 hours it's stable ~0.4ppm.
Overall delta from fresh built state to stabilized is -7 ppm, give or take one.

"Jump" between 06/08 and 01/09 is due to data was sampled in other files. Reference was continuously powered, without interruptions.

15MB DSV-file with RAW data, if you want to fiddle with it.

[Andreas]

Hello,

(I know I am repeating myself but)
those negative peaks of around -1.5 ppm at the end of the plot are most likely due to EMI-noise.
So I would place 2 * 100nF capacitors directly between pin 6+7 and 4+7 of the LTZ to reduce those peaks by around factor 10 before any other adjustments.

With best regards

Andreas

[Kleinstein]

One has to be careful with an additional capacitance between pins 4 and 7  - this can upset the voltage loop. It depends on the rest of the circuit if it can tolerate it.

There are other possible reasons for such peaks.

[MisterDiodes]

You are on the right track - That data set is very typical of LTZ operation as the substrate crystal lattice relaxes post-soldering over the first few thousand hours.  Let it run another year or so (we go 18 months minimum  burn in for customers) to see the real story - most units will see some more improvement. 

As noted by Kleinstein - Adding capacitors:  Definitely not recommended by LT apps engineering: that can cause instability in certain situations depending on board thermal flow.. If you need to do that find out where external noise is coming from. HP3458a Vref board / datasheet is an example of a very good, proven implementation over hundreds of thousands of boards. We have never seen any need to add caps, but your application might be different.

You probably have an external noise emitter somewhere near your lab - for instance when the AC compressor kicks in, or somebody turns on the microwave down the hall...or your neighbor across the street turns on their microwave or a water well pump kicks on somewhere in the vicinity of your lab.  Something like that.  We have tracked down those types of noise sources that show up on an LTZ especially if the traces are too long and current loop antenna area too large. Generally that's not the LTZ circuit per se.

[MisterDiodes]

Something else to look around for:  We tracked down one noise source similar to what you see as coming from a Weller soldering iron station that was left on in the next room.  It was on a different mains circuit but still was causing enough EMI pollution every time it cycled off to disrupt a ppm level measure - the glitch would appear only every once in a while depending on where the turn-off event happened during the power mains cycle.  Very random.  This was the type of iron that has the thermo-mag switch in the hand piece.

You have to be on the lookout for anything like that.

[tszaboo]

When I was doing PPM measurements, with a 3458A, I could tell, when the last person left the office down to the minute. And air conditioning added really a lot of noise on top of that. Also, I've tried to do the Tempco measurements when the device was cooling down. Because the heater is so random. Just insulate it, make part of the insulation adjustable, so it warms up or cools down in 8 hours, setup the measurement and leave.
And you can preatty much forget about measuring the tempco of a well built LTZ1000 circuit. You can maybe heat it up to 50 degrees,  based on the temp setpoint, and cool it down to 25. So I give them (management) a number, less than 0.3ppm/K. Or something. Buy me a better multimeter, then I will measure it for you.

[TiN]

I tend to agree on caps with MisterDiodes and Kleinstein. I tried them on one of my KX boards before, that didn't show any improvement. Best way to fix EMI/RFI issues is not to generate them, instead of battling with the consequences. However it's up to ManateeMafia, if he wishes to enter the murky waters...

Also one can see the temperature stability was not that great, due to use of portable AC unit in same room, as far I know.

So I give them (management) a number, less than 0.3ppm/K. Or something. Buy me a better multimeter, then I will measure it for you.

.
That's the good one, as there are not an awful lot of the voltmeters with tempco better than 0.3ppm/K  .

Tempco measurements on unverified/new LTZ1000-REF builds and designs however are useful, as a proof that design does not have some unexpected horrible terrible error or problem that give unexpectedly large TC. If tempco is buried in noise of own 3458A data on 20-30C span, that's good. If it's over 0.2ppm/K, then there is a problem that need fixing. This test data can be a good example of exactly that case.