Improving LM317 regulator Tempco drift

Status
Not open for further replies.

Mosaic

Well-Known Member
Hi All:
This is an empirical test I just did with on-hand parts to address using the LM317 as a good voltage regulator that has minimized temp. drift.
The spec sheet data indicates a 1% tempco drift. I had a circuit using 5.12V as a VCC reference for an ADC. However, I can get a drift up to 0.04V down as the LM317 warms. which impacts 2 decimal precision. This is made worse as I am measuring up to 20v and have a 4:1 voltage divider which quadruples the apparent precision error with any Vref drift.

Used 100°C hot air from 2" away from the TO-220 regulator to test this.

The adjustment resistor is setup as a 150Ω from Vout to the adjustment leg node feeding a 10K trimmer pot paralleling a 510Ω to ground. Adjusting the trimmer delivers the 5.12V. A 0.1uf cap sits across the trimmer for 'noise' control.

I have found by adding a 47K SMT thermistor (3960 beta) close to the LM317 (TO220) legs in parallel with the 150Ω resistor cuts the drift to .005V or by nearly an order of magnitude. This is a cheap solution as compared to a LM4040 4.1AIZ part to supply a solid Vref. It also saves a pin on my PIC ADC project, which is in short supply.

crutschow

Well-Known Member
I have found by adding a 47K SMT thermistor (3960 beta) close to the LM317 (TO220) legs in parallel with the 150Ω resistor cuts the drift to .005V or by nearly an order of magnitude. This is a cheap solution as compared to a LM4040 4.1AIZ part to supply a solid Vref.
But each unit can have a different amount of drift with temperature, so a single fix may not work with different devices.

A TL431A is an inexpensive precision reference which should have much less drift than the LM317.

ronsimpson

Well-Known Member
Have you looked at the "pot's" temp drift? I found many are not good.

crutschow

Well-Known Member
Have you looked at the "pot's" temp drift? I found many are not good.
That's true if you use the pot as a variable resistor.
If you use it as a potentiometer (voltage ratio) then the voltage will be much more stable, since then it's only the differential resistance drift between the top and bottom of the pot that will affect the voltage, which is likely much less than the absolute change.

Thus the circuit on the right below is likely more stable with temperature than the one on the left.
The disadvantage is the voltage adjustment for the one on the right is non-linear with rotation.

Mosaic

Well-Known Member
I did heat the pot directly...nothing noticeable driftwise from it . It is a Bourns Cermet type.
I looked at the TL431 as I see them in SMPS feedback loops, but this post cast doubts along with the lack of substantial power delivery.
http://forum.arduino.cc/index.php?topic=179769.0

Edit: Ok heated the pot as much as I dared w/o melting it....got a drift up of about 7mV But it is shunted wit an SMD 1% so that prob limits it's tempco. I think I can reduce the pots effect by shunting it with a smaller resistor and placing another fixed unit in series to make up the difference.
I used it as a shunted rheostat to reduce the current thru it.

Last edited:

crutschow

Well-Known Member
I looked at the TL431 as I see them in SMPS feedback loops, but this post cast doubts along with the lack of substantial power delivery.
I thought you were looking for an accurate reference voltage for your ADC.
Doesn't the ADC have a separate input for the reference voltage?

If not, then below is the LTspice simulation of an LM317 with a TL431 reference to control the output voltage, so that the regulator's output voltage drift is almost completely determined by the TL431 drift.
And using the pot in the potentiometer configuration, as shown, will largely eliminate the effect of its temperature coefficient also.

Attachments

• 993 bytes Views: 54
Last edited:

Mosaic

Well-Known Member
Well, I built the circuit and the results were disappointing. The performance was worse than the NTC compensated LM317. Plus it took a long time (10 min) to settle after being heated. I think the To92 case of the TL431 holds heat longer and its real world performance isn't up to what you'd want.

echo occurred.

Mikebits

Well-Known Member
I have some experience designing voltage references. I used a LTZ1000 for my premo reference. One thing I learned is that you can not design a temp stable reference on the cheap (inexpensively). Aside from the price of my LTZ part, the resistors were very costly. I had purchased Vishay S102 series resistors as they had pretty low tempco. The resistors are critical parts of the ref design. A fairly descent Volt ref part is the LT1021.

Mosaic

Well-Known Member
Yes, it does seem to take some effort. I customised the circuit a bit by adding a 4.7K in series with the 47K NTC (NTC close soldered to centre pin) and got the variance down to about 4mV peak with significant hot air on the LM317 and also tested by driving the LM317 to 2 Watts dissipation with no heatsink (105°C). With just the NTC alone it was over correcting a bit.
I'll live with that as it takes two low cost parts I already have in the BOM and very little PCB space. The NTC solution has to be customized to match the LM317 voltage setting resistors which are dependent on the desired voltage ref. Also the PCB layout and trace thickness matters.

Edit: that's about 800 ppm. A long ways off the LT1021 spec. of 100ppm, but about 1/10th the cost and works as a decent linear supply.

Last edited:

Mosaic

Well-Known Member
Looking at making a DIY device that can profile an LM317 to match it to the circuit...... With such an approach obtaining around 1 or 2 mV drift is possible at low cost. It will be a matter of the DIY circuit advising which resistor value is required after thermally cycling the LM317 to determine the 'trim' required for it.

schmitt trigger

Well-Known Member
I feel your pain...having had to design good performing circuits, within the constraints of a vanishing budget.

Your NTC solution is actually quite clever. As long as Crutschow mentions, you are aware that a single value may not be optimized for the tolerance spread found in normal manufacturing.

Mosaic

Well-Known Member
Quite so.
I have seen LM317 tempco all over the map. Which is probably why it's specified so poorly.
Hopefully there is a way around this which I can automate.
I'd like to be able to buy LM317s on sale (can get them for around $0.20 then) then run them thru the profiling or characterizing tempco gizmo, mark each with a silver sharpie with the resistor value that compensates the thermistor beta to trim the LM317 tempco drift to sub 2mV (over 85C range) and then all is good. The outcome is a decent variable vref/vreg for$0.50 incl the thermistor. Perhaps comparable to \$5.00 Vrefs.

schmitt trigger

Well-Known Member
Well, just remember that the NTC's curve is non linear.
You may be able to compensate for a narrow range, but not for the whole operating range.

Still, is an idea that I would at the very least investigate and validate.

Status
Not open for further replies.