TL431

[GigaJoe]

to be precise AZ431AZ-ATRE1 , Diodes Inc.

3 in parallel , current source on the same 431,  with 50ppm resistors , 10mA  for all 3 ; OP77 as averaging.
it in cardboard box, somewhere tossed far away ....  and occasional measurement

seems after 1 year result kinda acceptable,


result:

Month   Day  Voltage (Temp in C , approx)

++++++++++++++++YEAR 2022:
05 20 START UP

07 05 2.49237
07 06 2.49219
07 07 2.49207
07 11 2.49188
07 13 2.49173
07 15 2.49160
07 16 2.49155
07 17 2.49168 (21)
07 23 2.49171 (20)
07 26 2.49166 (22)

09 06 2.49345
09 27 2.50434

++++++++++++++++ YEAR 2023:
08 15 2.49612 (23)
08 18 2.49610
08 24 2.49611
08 27 2.49611
08 29 2.49609
09 04 2.49610
09 08 2.49613 (23)
09 20 2.49620 (19)

(last result , colder dry weather , vs hot raining days )

[armandine2]

I was playing with a TL431C last weekend --- beginner's stuff from H&H 3rd edition, the overvoltage crowbar with a triac

... did the rough Vref calculation using the values of the potential divider resistors and the Vin at the overvoltage point.

Came out as ~2.5V

which was surprising to me as I saw in H&H 2nd edition a Vref value of 2.75V 

[EC8010]

Typo. We all know that TL431 reference voltage is 2.5V. And I'm sure H&H do, too. You'd be amazed how easily typos creep in. TL431 is quite a useful part.

[armandine2]

they were consistent with this one

[EC8010]

Consistent error noted. I briefly wondered if TL431 reference voltage had changed through lifetime, but I've replaced a TL431 on an ancient piece of kit (late 80s) with a new one without needing major changes. Looks like an author error rather than typesetting. Easily done. I remember stating "Two squared is nine" at the whiteboard of my first lecture and it was only on the third instance of saying so realised what was causing the error in my calculations. Moral: At the whiteboard, simple arithmetic encounters quantum uncertainty, and students will swallow anything if it's spoken with enough conviction.

[Doctorandus_P]

Recently I was reading a bit in an old Ti design manual and noticed that a TL431 with the nominal 2.495V has the flattest output voltage over temperature. There is no such graph in the datasheet of the AZ431AZ-ATRE1, but I also looked at the Onsemi TL431CLPRPG.pdf and the graph of temperature versus output voltage on page 5 (See attachment) confirms this.

[Johnny B Good]

 That graph is suggesting that the tempco point of inflection is strongly dependent on the actual voltage output of each individual sample.

 I bought a pack of ten '431s on the strength of that graph with the intention of selecting one with a 2.500v value to place the tempco point of inflection closer to the "sweet spot" of the 45 to 50 deg C at the internal trimpot location in my LPRO 101 Rubidium oscillator.

 Although the existing 5K trimpot is a recent replacement of the original worn out trimpot, I suspect it still might well be a source of instability I'd very much like to eliminate now that I'm in 10 E -13 stability territory with my temperature stabilised, barometrically compensated RFS project.

 Currently I'm relying on this coarse internal adjustment to preset the external 10 turn calibration pot to its midpoint (a ten turn 5K helipot connected to a dedicated temperature stabilised μ7805 voltage regulator with the wiper connected to the C field input via a 3.3M ohm resistor acting in conjunction with the 150K input impedance as a 23:1 reduction of the 0 to 5v range of adjustment in lieu of the more problematical use of padding resistors either end of the pot).

 The only downside to making this modification being the loss of the coarse internal adjustment to conveniently preset the external pot to its mid position. However, I can live with this since I can simply wire a fixed selected value resistor between either the 5 or the 0v to the C field pin to preset the cal pot to its mid point. The TL431 should offer better reliability than a trimpot (or at the very least prove whether my concerns over the stability of the trimpot were justified - I can always reverse the mod if I deem the exercise a failure).

 I haven't done the mod yet since the bench is currently occupied with my MK III gpsdo project and I'd like to see just how close to the nominal 2,5v mid point setting that the internal pot is actually set to when the external cal pot is at its midway setting before I make any changes.

[Johnny B Good]

UPDATE!

 I finally took the plunge last week and replaced the trimpot with a TL431 (2.51235v) which shouldn't be too far off its tempco inflection point. I simply mounted it so it sits tensioned against the upturned edge of the base plate to hold it some 2 or 3 degrees above the heat spreader temperature (36.050 degrees +/- 10mK), hopefully, stabilised somewhere between 38 and 39 degrees.

 After doing a test with a 1.5v test voltage across the pot, I'd determined that it had been set to output 3.2 volts or so. A little further away from the hoped for 2.5v mid point than I'd have like but since the thermomechanical induced errors and the rather poor adjustment repeatability setting specifications would be banished forever, I decided to carry on with the mod knowing that I'd now be relying on resetting the external calibraton to its mid point using a selected fixed resistor value to, in this case, pull the frequency adjust pin down towards the ground rail potential.

 After doing 'some math' I chose a suitable resistor to test with, followed by more math to refine the resistance value yet further before 'permanently' mounting the extra resistor against the heatspreader to run an overnight test. which revealed a large frequency swing on account I'd overlooked the fact that the benchmeter had been adding its own 10M to the additional 10M I'd strapped across the 820K resistor I'd been using.

 I'd switched the bench meter off before I went to my bed which isolates its input from the outside world, thus disconnecting that additional 10M shunt impedance from across the circuit, So yet a third go at finalising the external compensation resistor network. I landed up using a 280K (made up with a pair of 560K 'cos my book of 1208 1% resistors had a strip of 910Ks where the 270Ks should have been placed, I also replaced the 3.3M padding attenuator resistor with a 1M in order to double the tuning span and compensate for the reduction from 150K at the tuning pin down to 100K.

 All in all, a rather more complicated job than I'd hoped for. At least, If I need to readjust the external pull down resistor value at a later date, I won't have to open the LPRO-101 again to fiddle with its innards.

 Initial tests were promising in that there now seemed to be less startup drift post base plate stabilising at the 36.05 degrees set temperature. Reassuringly, the time from stone cold power up off a 19v laptop charging brick to atomic lock was its normal 191 (+/-1) seconds. At least now, I don't have to contend with the stochastic effects from a crappy internal trimpot (the range of adjustment offered by the external frequency trim adjustment alone is way more than ample for my foreseeable needs).

 I no longer find myself having to make random re calibration adjustments to keep up with the internal trimpot's lack of precision and stability which had been the whole point of this exercise in the first place.

 BTW, I've attached three screenshots of the Arduino's serial plotter output of baseplate temperature values (about 6.4 seconds' worth) showing a total of 500 points, including those re-scale defeating 'spikes' inserted every 498th point (a useful side effect that allows a stopwatch to be used to measure the control loop cycling speed - currently 78Hz). The display is typical over an ambient range of 14 (or even lower see the footnote) to 27 30 degrees. The Y axis is scaled in mK with 0 corresponding to the original 36.000 deg baseplate datum temperature point.

 NOTE:  I've not had a chance to re-verify the low ambient temperature tolerance test result down to 4 deg that I got before fixing a serious bug in my temperature regulation algorithm.