TL431 temperature compensation method question

Status
Not open for further replies.

d123

Hi,

Please no snarky remarks about the ridiculously unrealistic 33.98251k resistor, only purpose was to get exactly 10.000V in the simulation.

I can't re-create 0ºC to 100ºC here at home nor have appropriate tools to measure the effects on the DUT anyway.

Question: Does simulation result "B" look credible?

I possibly more or less understand the premise of what the diodes do but no idea why it turns the expected "banana" into a "sideways s" either. Diode leakage increases with temperature and draws current away from ref pin, which internally in the TL431 is an NPN base.

Checked without feedback and that has no effect on temperature performance, it's apparently only with the feedback resistors that there is a small improvement in performance across temperature range. One diode helped a little, two a bit more, three spoils the effect. 1N4148 works, 1N4007 was no use and Schottky is pointless.

Thanks.

Easy peasy

A better reference required ...

d123

d123

points: 2

FvM

Super Moderator
Staff member
I fear, it's an error of reasoning to expect correct modelling of actual hardware temperature drift with your simulation circuit.

There's no clear rule how simulation models account for error quantities like offset voltages and currents or temperature drift. You need to be careful with conclusions for your circuit design.

The only trustworthy information source is the data sheet. According to it, there's no systematic reference voltage drift that could be compensated by a pre-calculated circuit. We see however a typical drift of reference input current, suggesting that the reference voltage divider should rather have 1k than 25k impedance not to cause additional drift. Your simulation results suggests that Iref temperature dependency is not correctly represented by your TL431 model. You can check the SPICE subcircuit in this regard, or measure the current in the simulation.

Using 1N4148 reverse current for compensation purposes has to deal with a great type variation of this parameter, I also doubt that it's a good idea to compensate a linear or quadratic temperature dependency by an exponential temperature dependency.

d123

d123

points: 2

wwfeldman

I can't re-create 0ºC to 100ºC here at home nor have appropriate tools to measure the effects on the DUT anyway.
use a cardboard box, an incandescent bulb, a temperature sensor and a control loop of your choice

i did this some years back to demonstrate open loop, closed loop and hysteresis
to maintain temperature with a 60 watt bulb and a cylindrical oatmeal box - it has a well fitting lid

d123

d123

points: 2

asdf44

Yeah, not too hard to find a better reference.

And don't forget the resistor drift.

It's better in terms of cost/performance to get a fixed voltage reference than to buy low PPM resistors (unless you really need adjustment).

d123

d123

points: 2

d123

Thanks. Very, very informative replies.

Nice idea about the bulb for a heat testbox, super.

Post #3 Very eye-opening points made regarding simulation limitations, actual devices and also last sentence pointing out mismatch between linear/quadratic/exponential "problem" and intended solution, many thanks.

#2 and #5 comments lead to question: For a circuit that would only be used intermittently for a few minutes at a time between 10ºC to 40ºC ambient, is it bad design to stack two voltage references such as two grade D (that D stands for "dismal choice by user", doesn't it?) 5V LM4040s to get 10V? Bearing in mind that that's all one has in the cupboard reference-wise, besides a few LM4041 1.2Vs. I'd have thought the TL431AC would be the best choice to make a 10V ref in this case with +-100ppm/ºC resistors.

Easy peasy

depends onthe accuracy you require over temp - also how stable is the volt source you are running the ref from? ) over temp..? )

d123

Hi,

depends onthe accuracy you require over temp - also how stable is the volt source you are running the ref from? ) over temp..? )
Something like the REF102 (10V +-2.5mV) would be delightful, but returning to Planet Earth... I would like 10.000V at ~20ºC ambient, but again - not realistic. 9.990 to 10.010 would be good, allowing for ~ +-3mV ripple. Not sure if I'm overstretching ability and/or reality here.

Hadn't thought of voltage source stability. Have had to go from 1V ref/Voltage source to 10V ref/V source idea, as error will be less pronounced and easier to measure with a degree of confidence it's accurate, based on Saturday's demoralising breadboard test run of 1V +-10mV.
Was going to work off 4*1.5V AA batteries and LDO Vreg TPS73801 SOT-23 (it's good enough for my needs, could stretch to a Micrel 29502 TO-220 but seems a waste of a very good LDO on this low current circuit).
Supply will have to be a 230V mains-based 230AC to dual output 15VAC encapsulated transformer into the regulator (bridge rectifier and smoothing caps before the regulator, obviously), followed by the 10V reference. I'd considered Vreg 13 - 14V > Vref 12V > Vref 10V but I'm under the impression that is overkill/redundancy in action/using an extra reference to only achieve the same amount of buffering and voltage stability and the same temperature drift - isn't it?

I've seen a problem with stacking two 5V LM4040s - according to two different DMMs I have, it outputs 5.01V on a 10k load (actual load would be 1G + 10M in series) - two would make 10.02V...

Set a TL431 to 5.07V and compared to LM4040 5V and LM4040 seemed more stable (rarely moved to 5.02V from 5.01V) whereas 431 appeared to move up and down 10mV and sometimes even 20mV when blowing on them or cooling them a little. I assume that is the slow feedback it has or my (not home-made) DMM doing too many readings per second which in the end is somewhat unhelpful perhaps.
One thing is that the power supply on the cheap oscilloscope seems a bit poor - 5V is always ~4.76V and 12V is ~11.84, on tiny loads of a few mA (best not actually try to get the advertised 1A out of it!) and its own osc probe says it has a puzzling 130mV ripple.

To be honest, I think I may "risk it" with a TL431 set to 10V as I'd have more control over the output voltage and for e.g. maximum 5 mins use at a time (and allowing circuit to cool down if need be - this is for home use, not on an industrial scale 24 hours a day - it should be able to show me what I want to read about a resistor value to use it in a timing application to measure 10pF to 1nF capacitors.

I don't understand why it is considered a "low quality" reference when it's used so much. Although comparing $0.25/TL431 to$8/REF102 may be a good indication as to why you and asdf44 have said "change to a better ref".

Easy peasy

If you have to step up froma 2v5 ref to 10V the % error is the same - only an intrinsic 10V ref would be able to give better stability ...

I think the "best" TL431 you can buy is 0.5% ? are there much better versions?

d123

If you have to step up froma 2v5 ref to 10V the % error is the same - only an intrinsic 10V ref would be able to give better stability ...
The more I think of it, the more tempted I am to wait until I can afford the luxury of a 10V REF102, after getting endless bills and pressing needs/priorities out of the way... Anyway, it would make a critical/vital part of the circuit not worrisome/leading to doubts at each power-up and also not needing re-calibrating quite so often; and most importanltly, far more accurate + stable than I can hope to make with what I have to hand at present. I suspect it will be worth the wait and there are plenty of other parts of the two circuits to consider properly beyond the breadboard rough sketches already done to check simulations have more than a good semblance to reality.

I think the "best" TL431 you can buy is 0.5% ? are there much better versions?
I am under no obligation to confirm nor deny this allegation. As you know, standard is 2%, A is 1%, B is 0.5%, but there's no C grade as far as I'm aware.

Thanks for helping lots with thinking this part of the circuit through properly.

Easy peasy

I am not sure there are any intrinsic 10V references out there ( or even 5V ) I should know as we do a magnet power supply to 10ppm, 100A 4 quadrant & 600A 2 quadrant ... so just pick the best reference for the money and go from there ...

d123

Hi,

I am not sure there are any intrinsic 10V references out there
Oh yes there is, if you're interested, REF102C, and it's not only from the days of Burr-Brown but it's also active and just utterly super looking. Seems better than some of the floating gate ones I've read about. Mentioned it earlier on: 10V +-2.5mV, +-2.5ppm/ºC. datasheet and manufacturer product page. 102A is +-10mV, 10ppm/ºC. Price could be an issue but for a one-off where it is the "heart" of the circuit, seems worth the price of a pizza.

I don't even know what a magnet power supply is. Those are incredible parameters, impressed.

Easy peasy

Easy peasy

points: 2

Easy peasy

Well done - I see it is based on an 8v2 buried zener reference - I may well use it myself.

Staff member
d123

d123

points: 2

KlausST

Super Moderator
Staff member
Hi,

you expect high accuracy.
Assembling, soldering, mechanical stress, thermocouple effects and so on may degrade (besides other devices) the overall peformance.

The datasheet tells you the performance of the reference IC... but carefully read the test conditions.
Also be sure to carefully design the schematic and the PCB layout not to add other errors.

Klaus

d123

d123

points: 2

Easy peasy

hello Brad, a 4v7 and a 1N4448 in series gives surprisingly good performance for a ref - with a const current drive .... not 10ppm though, more like 0.2 %

points: 2

asdf44

Yeah to echo what Klauss said where is this reference going?

Every part it touches needs to be in the 10ppm territory if this ref is going to help you. That means expensive resistors or matched resistor packs, well chosen opamps, ADCs with low gain drift (ADC's have gain issues independent of reference), the right topologies (differential amplifiers, remote sense) effective RFI filtering, etc etc.

d123

d123

points: 2

Easy peasy

Yes, good shielding often necessary - i.e. the tinned Cu box, all supply lines need good filtering ... etc ...

d123

d123

points: 2

d123

Hi,

... Problem with phone line at present so won't be around for a while...

Thanks for all the additional info./advice, helps with working through circuit as carefully as I can. Goal is to minimise errors I foresee as far as I can realistically in my case, am aware of many - not all - of problems mentioned in last posts. A long way to go with this one...

Thanks!

Status
Not open for further replies.