Is tl431 good for reference in 19 999 counts voltmeter? Any variants?

[001]

Is tl431 good for reference in 19 999 counts voltmeter? Any variants (NOT LTZ1000 please)?

[T3sl4co1l]

What does the datasheet say?

How many of those counts are supposed to be accurate?  All of them (i.e., noise and/or precision <= 1LSB)?

Tim

[001]

I`m confused becouse I see it in industrial-made unit

[T3sl4co1l]

Hey, just because I pick my nose, doesn't mean I'm right...

Tim

[floobydust]

You have to weight cost, accuracy, noise and temperature coefficient, as a few parameters.

I don't consider TL431 to be a great choice. Quality is all over the place, and I find them noisy - but they are the cheapest part in town at $0.50
Why did you spend $10 on the A/D to spend $0.50 on the reference IC and $0.001 on the divider resistors? This example means you need to balance your design.
Look at Linear Tech AN-82 Understanding and Applying Voltage References

For cheap I use LM4040 worst case ±100ppm/°C and typically ±15ppm/°C for $1.00 and 35uV RMS noise.

Better parts cost several dollars.

edit: kinda discussed already https://www.eevblog.com/forum/projects/low-cost-voltage-reference-experiment/
added LT app note.

[Wolfgang]

IMHO, the unit that uses this thing for a 4 1/2 Digit display is crap from China using an 7135 or similar IC.

Reasons:
- The best TL431 grade is 0.5% accurate
- TK is not spectacular, either
- No drift specs at all.

Recommendation:
- LM399. A few bucks, but stable as a rock and low tempco. LTZ1000 is only needed for top end stuff.
- If you need to be cheap: LM723. Low tempco, drift spec available.

[Kleinstein]

Depending on the manufacturer there are TC and accuracy specs. Absolute accuracy of the reference is usually not an issue anyway as the voltmeter usually would need an adjustment anyway.

For cheap ADC chips the 2.5 V level is not that bad - here the classical 7 V zener refs like LM329, LM399, 1N829 are more like a problem and not really suitable.

The noise of the TL431 can be low enough, but there are better options. Stability of the reference and the resolution of the DMM are not necessary linked. So even 6 digits with a lousy, high drift reference is possible in principle, though usually not a good idea.

The TL431 is not even better than ADC internal refs sometimes found.
Instead of something like ICL7135 and TL431, I would consider something like MCP3421 with the internal reference.

[Wolfgang]

@Kleinstein

did you find a TL431 with long term drift specs anywhere ? Not even the TL431BI from TI has one.
And a reference without a drift spec is not usable in a meter, IMHO.

Regards
  Wolfgang

[001]

4 1/2 Digit display is crap from China using an 7135

Thanx for Yours opinions!

Is a 7135 a real crap?
or
Did You mean what UNIT is crap if it based on 7135 WITH 431 reference?


I see what 7135 is more stable then tl431 and looks adequate for 4.5 digit purposes like 0.05% meters
It needs better reference when 431 IMHO

[Wolfgang]

... sorry for my lack of precision. The 7135 is fairly OK, old but good enough for 4 1/2 digits.
The combination with a TL431 is simply nonsense, I have only seen them in cheap Chinese panel meters.

[001]

If You say about 7135:
It requires reference about 1.2V
But LM*** voltage are fare high. Can I get better result in 1.2V range with LM*** and resistor divider since resistors drift is headache?

[Wolfgang]

The INTERSIL datasheet suggests an ICL8069 reference plus a multi-turn precision pot to make exactly 1V.

IMHO, you may use any *good* reference as long as you have a ratiometric divider for the IC ref input voltage.
An LM4040-2.0 would be OK, it has a long term drift spec of 100ppm/1kHour. Tempco is (typ) 15ppm/°C, precision 0.1%.

[GigaJoe]

personally, i hate TL431 
why no one mention LM336-2.5

[001]

why no one mention LM336-2.5

Why?

[Kleinstein]

If You say about 7135:
It requires reference about 1.2V
But LM*** voltage are fare high. Can I get better result in 1.2V range with LM*** and resistor divider since resistors drift is headache?

If one needs 1 V (e.g. ICL7135) starting from 1.2 V has some advantage, as resistors are only responsible for the 0.2 V drop and thus only about 1/5 of the voltage, compared to nearly all if one starts with a 7 V reference. So resistor TC is less critical by about a factor of 5.

The other good starting point would be a 2 V reference and use a divider with 2 identical resistors (e.g. 15-50 ppm/K thin film type) and hope for good matching. A pot (possible digital pot) would than be only for fine tune (e.g. +-2%). AFIAK there are a few ready made 1 V references, but one would still need some range for adjustment if a display oriented ADC is used.

Given a 4 digit resolution and maybe a 5-10 K span for the temperature is would be nice to have a TC in the 10 ppm range or better. For the ranges other than 2 V (or whatever is the primary one without gain) one would be limited by the resistive divider anyway.

[001]

is would be nice to have a TC in the 10 ppm range or better.

Any common IC? Not exotic

Thanx!

[Kleinstein]

The LM4140-1 would be a nice option.  1.024 V could be easy trimmed down to exactly 1 V with little effect of the divider.
Though it is not that cheap (e.g. $2  - 5 depending on the Grade / TC).

However with such parts, it could be better to check at your favorite source and than look what is available there.
There are lots of 1.2x V references(usually bandgap - which is OK here) to choose from. The part directly competing with the TL431 would be an LM385 - similar performance and price, but more suitable 1.2x V.

[David Hess]

TL431 type references are not specified for this type of application.  They could be as bad as 150ppm/C which works out to 3 counts out of 20,000.  There are much better "cheap" parts like the LM385 which could be 25ppm/C.

[floobydust]

ICL7135 is usually paired with old 1N829a $15 or ICL8069 (obsolete) reference for 50ppm/°C
I think a 5°C change in temperature then results in 250ppm or 15 counts.

I looked at the under $2.50 league (in low quantities) and across the board you get 20ppm/°C.
It seems to be the per-part trimming that drives price up.

MAX6070 6-8ppm/°C  looks like the best (datasheet) performer. I guess Maxim is trying to get their foot in the door. But unknown trimming tech and long-term drift graphs missing from the datasheet Fig. 4,5,6. 

ADR3412 (I think used in EEVblog GW121) 8ppm/°C but $2.80 and no noise cap pin. Uses their DigiTrim (blowing polysilicon fuses).

I don't think EEPROM trimming is any good for the long term.

[splin]

You have to weight cost, accuracy, noise and temperature coefficient, as a few parameters.

I don't consider TL431 to be a great choice. Quality is all over the place, and I find them noisy - but they are the cheapest part in town at $0.50

For cheap I use LM4040 worst case ±100ppm/°C and typically ±15ppm/°C for $1.00 and 35uV RMS noise.

Better parts cost several dollars.

Diodes Inc do various TL431 clones. The AP431IAZTR-G1 has better -40 to +85C maximum temperature coefficent of 4000ppm compared to 6500ppm for the LM4040A (100ppm/C max average). Over 0-70C it's even better with a max deviation of 6mV or 2400ppm V 4500ppm for the LM4040A.

Better still it's approx 1/10 the cost at less than 5 cents (qty 100+, Avnet). The noise isn't specified so probably not good. Hysteresis and long term drift are not specified either but the LM4040 specs (typical only) are very undemanding so I would be surprised if the AP431 was any worse.

The LM4040A initial tolerance is .1% compared to .5% for the AP431 which will important in some cases (but the latter is slightly better overall, initial + temperature deviation -40 to 85C). Are there any other significant benefits to the LM4040 to justify the much higher price?

[001]

Hi
Thanx for Yours opinions!

$6 chip will be ok too

[floobydust]

You have to be careful comparing the same part number between manufacturers. It just means most specs are equivalent, some worse, some better, some omitted.

The AP431 seems obsolete, replaced by AP431S- which still has no aging, noise specs.
Diodes Inc. parts have been trouble for me in production, I no longer use them. They are cheap for a reason.

The TL431 (die pics and some reverse eng) has different design priorities compared to a precision reference.
It seems to have higher gain, higher shunt current range, lower accuracy and no aging or noise spec. - it is better suited for regulator applications IMHO.

Cheaper voltage reference IC's are using EEPROM-registers for trimming, instead of laser-trimming.
On-Semi LM4040, LM4041 "... ON Semiconductor’s  Charge Programmable floating gate technology ensures precise voltage settings offering five grades of initial accuracy; from 0.1% to 2%" {sounds like eeprom to me, like in TI LM41xx}
TI LM4040 "... uses a fuse and Zener-zap reverse breakdown voltage trim during wafer sort to ensure that the prime parts have an accuracy of better than ±0.1% (A grade) at 25°C"  But even these have changed with time.

Assuming OP has a trimpot for A/D calibration, we don't care about initial accuracy, just stability.

If ICL7135 is fed 1.0000VDC ref, I wonder if going up to say $6 is worth it, compared to the resistor's trimpot/voltage divider drift and tempco.

[001]

If ICL7135 is fed 1.0000VDC ref, I wonder if going up to say $6 is worth it, compared to the resistor's trimpot/voltage divider drift and tempco.

Can You say few words about it?

[Kleinstein]

There are other error source, than just the reference. This is especially the divider at the input or amplifier for ranges other than 2 V.
Also the divider / trimmer to bring the reference to 1V adds to the TC / aging.

The LM4040 / LM4041 / max6070 and ADR3412 look good. Depending on the source and grade they seem to be in the $2-6 range.

Starting with something like 1.2 or even 1.024 V reduces the effect of the divider. So a 2.5 V reference like the TL431 or like is a poor idea to start with. The natural voltage of a band-gap reference is 1.23 V. It does not make sense to save on a cheap 2.5 V reference if one would need to spend the money on a better divider (e.g. 2 PTF56 type resistors for 2-3 EUR each).