5V Precision Voltage Reference

[Bootalito]

Thanks.  I was reading that the LM317 needs about 3.5-10mA minimum, so I think I'll go back to the LM431

[Bootalito]

Yes the LM317 needs 2-10mA minimum, so that's not going to work.  I think I'm going to switch back to the LM431.  This will reduce my run time to a measly 16h. 
I researched the T1763 and I think this is a really good choice for the next version.  I'm just trying to get my feet wet right now so I can't justify a $5 linear reg just right now.

[David Hess]

Thanks.  I was reading that the LM317 needs about 3.5-10mA minimum, so I think I'll go back to the LM431

Or add a non-inverting low power operational amplifier powered directly from the input which uses the output as a reference to power the reference itself.  The LT1021 draws about 0.8 milliamps and there are plenty of operational amplifiers which draw less than that.

Or an even simpler solution is to take advantage of the minimum voltage drop across the LT1021.  Add an n-channel depletion mode MOSFEET or JFET with a sufficiently high gate threshold voltage in series with the input with the gate connected to the output as shown in this example from Maxim:

https://www.maximintegrated.com/en/app-notes/index.mvp/id/482

[Bootalito]

Ok so here is my "final design", unless someone sees any glaring errors.
I added an input bypass cap to the ref IC.  Re-added the LM431.  Got rid of a couple 100uF caps for the battery, as these are only going to be useful when I have this hooked up to an external supply for extra ripple rejection.  Added some solid regions for heat disapation through the board near the LM431 and passives.  Added cutouts to prevent the heat from traveling too far.

[Andreas]

Hello,

for me only R3 does not fit to the 15V regulated voltage.
At around 4 mA minimum load (1.8 mA for reference and resistors + 1.8 mA for TL431 + resistors)
 there is not enough headroom for the 1K resistor at the minimum battery voltage, so the TL431 falls out of regulation.

I would go for a 10V regulated voltage.
In this case it should be possible to increase R3 to 1.2 or 1.3 k without falling out of regulation at the minimum input voltage.
But be aware that you need up to 7.5 mA at the maximum battery voltage.

With best regards

Andreas

[Bootalito]

Yes, you're right, the NiMh could get down to 14V or lower.  I changed the output of the TL431to a 10Vnom and upped R3 to 1k.  PCBs ordered.  I'l take some pics in a couple weeks.  Unfortunately I don't have any good multimeters (BM869,  5 4/5digit, 500,000 count is best I have).  I'm going to do some basic tests, burn in for a month, then measure the reference at one of my customer's metrology lab.

I already have some grand plans to make an LM399 w/ LT1001AC similar to the app note w/ with some 18650 batteries.  But we'll see how far down the rabbit hole I go, especially with my limited access to good metrology grade measurement equipment

Latest version in case anyone cares to use: https://easyeda.com/terryjmyers/Precision_Voltage_Reference_copy-d601e8ca03ab409a8e9dfcc9fd1675d2

[Bootalito]

How do you create one of these powered from a SMPS, or a battery with a boost converter?  Would one add some gigantic caps(like 100uF?) for filtering, then feed that into an LT1763? Is that enough to get the ripple out?Would you create a PCB with a ground plane for the input switching side, then create a star ground for the output side starting with the LT1763, with a star ground right to the negative battery terminal/DC input jack?

I could also use some help explaining some readings I was taking last night as I was messing around with the LM431 that I'm going to use for the above project.
All voltages measured with a BM879 (on 500,000 count mode), as I don't yet own an oscilloscope(*ducks)
Input voltage from a linear power supply (Chinese grade) = 15.0290 + 2LSD, using crest mode: 15.20V max, 14.81V min! using mV AC = ~60mV.  hZ measurement = 0.0000
Output from LM431 = 10.7818V + 2LSD,  using crest mode: 10.97V max, 10.58V min! using mV AC = ~60mV (identical to input), hZ measurement = 0.0000
Placing a gigantic capacitor on the input (think it was 400uF) didn't make any difference.  Am I correct in assuming that if I had an occiliscope it would show an ripple voltage of 60mV RMS?!  Why wouldn't a capacitor smooth this out any amount?

[The Soulman]

I can't comment on your measurements but I'm pretty sure a oscilloscope wont help with measuring close to dc noise at such low levels.
Would like to suggest a alternative method on the assumption you have two references, ideally with two separate power supplies.
Connect the grounds of the references together and measure the dc voltage between them with the lowest range on your meter, if you have a lot of "flicker" then you have low frequency noise, if the values move steadily over time you don't have significant noise and you can use your references as a thermometer. 

I measure my lt-1021dmH-10's this way with a cheapo data logging meter with 10uV (1 ppm  ) resolution.

Ps,  lm7812 works fine.

[beanflying]

What is the voltage to Earth potential of your power supply compared to virtual ground measured on the AC range on your meter? Some of them float 80+ VAC above earth. Try this fix method to bring it under control if it is high https://www.eevblog.com/forum/testgear/feeltech-fy6600-60mhz-2-ch-vco-function-arbitrary-waveform-signal-generator/msg1344167/#msg1344167

Just throwing caps at unknown supplies and ripple won't necessarily help. Crack it open and see what is there for a start. It can be done by trial and error but the easy way is a different supply or run on batteries. My Bench cheap chinese 12V 20A fixed is under 20mV (switchmode) and my Manson switchmode runs under 5mV so switchmode isn't a problem but bad power supply design is.

[Bootalito]

BUILD COMPLETED:

I don't have better than a 5 4/5digit multimeter, but I adjusted the pot to as close as possible.  I also really don't have the means to characterize it easily myself. I'll take it to a customers metrology lab to get an initial reading within a week or so, then again after I age it for 1000h(I don't want to impose on my customer).  I'll be aging it with a switched mode power supply.  I bought some 600mAH Li-ION 9V batteries, and I recalculated a 160h run time.

Some initial data:
Circuit Power Consumption:
30.0Vin (C2 limiting at 35V rating) = 18.97mA = 0.569W
16.40Vin(4.1V/cell) = 5.33mA = 0.0874W
14.8Vin (3.7V/cell)  = 3.74mA = 0.0553W
14.0Vin (3.5V/cell)  = 2.95mA = 0.0413W
12.5Vin (minimum) = 1.89mA = 0.0236W

10.82V TL432 measured Vout
5.00000 +- 0.00010V Vreg Out. (I don't really have a good way to measure true noise, but 0.00010 was the total range it moved while testing.

main Zener resistor (1k 1W) dissapated 0.37W at 30V.  Gets a bit warm, but not overly, and I don't think it affected the voltage reg due to dissapating power through a large copper plane and the isolation slots.
12.5V minimum Battery voltage required before TL432 voltage started to dip.


Maximum Battery life estimation
    Average Power = 55.3mW @ 3.7V
    159h (6.6days) @ 8.8Wh (2x600mAh 9V lithium-ion)

Lessons learned for future builds or other people who read this in the future:
I have no way to easily trim it without having the IC exposed.  Should have used a side mounted screw trim pot and place near edge of board to drill a little hole in the box
I have no way to power it externally for the burn in period.  Should have added a DC power jack or exposed pluggable terminal blocks., and a little mosfet to switch between external and battery powered.  Maybe add a little charging circuit
Permenant marker doesn't write well on silk screen, so the large rectangle I added to write the final voltage on in a month is not very useful
Should have added some test points for a multimeter for initial build/testing.  Perhaps some exposed vias to stick a probe into.  One at the IC Vin, One at the IC gnd, One on the output.
My build was based around a box I already had, but I should have used a bigger box be able to position the banana plugs better, and not be so restricted in component placement.
I should have used a switch that more readily available.  I used a very high quality switch because I have a sample box of switches from a manufacturer.
The TL432 and SS43 didn't disapate as much heat as anticipated.  Would have made R3 copper pad much larger to dissapate heat.
I probably should have used a low noise LDO regulator like others had said.  I haven't measured the noise, but I'm sure it would have been better, and less power consumption by an order of magnitude




I cut the legs off of the NC pins.  I pushed the leads in slightly so they weren't under stress when sitting there before soldering (IC was loose in the holes).  I use a high flow rate 120mm fan to blow fumes accross my desk while soldering.  I moved this fan 6" away from the IC while soldering it to keep it cool.  Soldered one pin at a time at 280C as quickly as possible and let the whole thing cool off about 20sec before soldering the next pin (the fan cooled the IC off pretty damn quickly)


"Like a gloooove"



The plastic was too thick for my nibbler and my uh...dremel skills need some work, so the final build looks like crap...damn it...  Onto version 2!

[The Soulman]

  Kudos for getting it done.
For V2 you could have a second footprint for the "H" version.