Odd voltage reference

[MacGyverS2000]

As with all of my designs, I need a solution that's minimal in cost ;-)

I require a voltage reference (actually, I don't need the reference voltage itself, I'm only using the ref's go/no-go logic to determine if the battery level has dropped too much) but do not yet know the specific level... it will be somewhere in the 7-8V range, though. An acceptable piece-to-piece error would be in the neighborhood of 0.05-0.10V (+/-). This is a battery-operated circuit, so minimal current draw is of importance.

I've come up with two potential solutions, but I'd like everyone's feedback on their accuracy and repeatability. The first is to use a resistor in the ground leg of a reference diode (such as National's LMx85 series)... they list a typical operating current of 20uA, so a several hundred k range resistor would raise the operating voltage to the level of interest. My question to this design would be... would this be an acceptable design practice? Using an SOT-23 package, the total solution would cost around $0.40, well within my budget, and current draw is in the 20uA range.

The second design I thought of includes a FET controlling a resistor dividor network and an A/D channel reading the network every once in a while... if the network voltage drops below a certain level, battery is getting low. This circuit has the advantage of being less expensive (in the $0.10 range?), and may be the way I go. I'll have to run the quick calcs, but even 5% tolerance resistors should be accurate enough. It uses up an extra A/D channel, but I have extras, and since it's FET switched, I can turn it off while the processor is in sleep mode.


Design #2 looks like the way to go, but I still wouldn't mind any comments you guys might have...

 

[logbook]

For "design" #1, it looks like a non-starter. Let’s consider the LM385-2.5 and suppose you were just going to use it at 2.5V rather than 7.5V. Looks at the reverse characteristic of the device. It is just like a zener diode but with low current operation and a low dynamic resistance. Below 2.5V it has a very soggy characteristic. The typical curve shows 10µA at 2.5V but 3µA at 1V. This is not the sort of switching characteristic you wanted. And remember that this is not a characterised region for this part so the curve shape is not deterministic. Forget this circuit and pretend it was someone elses idea!

I don’t get how #2 is so cheap, given that you are using a FET to switch the top end of the resistive divider.

 

[MacGyverS2000]

The FET will cost me a handful of pennies, the resistors will be a penny or two... maybe not $0.10, but still quite cheap.

 

[felixc]

Why do you use a FET to enable the resistors. Since you don't care about the speed in this case, use resistors high enough and be within the range of the max leakage current. You want precision? Is there an EEPROM in your design? Create a "calibration" value to compensate your readings.
When you calculate your pennies, don't forget that the cost of assemly, per part, can be in the ten cent range. More than many passives.

 

[logbook]

Felixc’s idea may not be workable. It is all very well compensating for the effects of leakage, but since the leakage error may double every ten degrees C, a fixed calibration offset is not going to be useful.

 

[sreid]

Are there not battery management IC that would contain this function?

 

[MacGyverS2000]

felix, since the A/D needs to see a fairly low impedance (in the 2.5k range), the dividor resistors need to remain fairly small (two 5k look good). The current draw beng what it is for such value resistors, the FET is there to completely disable the network... I'll only enable it when it's convenient to, and disable it whenever the processor goes to sleep.

I think 5% is within my level of precision, but even if that turns out to not be the case, switching to 1% isn't going to cost me much more. As far as assembly costs, we're talking a penny a pad.

 

[ScottyUK]

Hi MacGyver,

You might discount this on cost grounds, although in space requirement it is similar to that of the FET idea: use the high resistance divider network proposed by FelixC, and buffer it using a micropower op-amp so the A-D doesn't load the divider. There are a number of micropower op-amps with a shutdown feature that drops the power consumption to virtually nothing when it is not required.



----------------------------------

If we learn from our mistakes,
I'm getting a great education!

 

[logbook]

MacGyverS2000

I suppose you realise that your numbers don’t add up. You said that you were willing to accept ±0.1V tolerance on your approximately 7.5V power rail. That is a tolerance of ±1.3%. Using two equal ±5% resistors will give you a tolerance of ±5%. Thus you should start with ±1% resistors.

What you could also do is divide the signal down less. If the ADC has a full range input of 5V for example you could divide the signal down by say 0.6. This will improve the accuracy (slightly).

 

[Skogsgurra]

A spare gate and a Mohm voltage divider often makes a fair enough voltage supeviser. Have a look at 74HCTxx thresholds. They are quite well defined in practice even if the guaranteed values have a wide spread. I do not say that this is a solution that works for you - but it has worked for me.

Zeners can sometimes be used but you often need lower voltages. Ordinary LEDs have threshold voltages from around 1.2 V (IR diodes) to 2.6 V for blue-green LEDs with 1.6 - 1.8 for red and green LEDs i between.

The green LEDs seem to have the sharpest knees with useful current levels between 10 and 50 microamperes. The voltage is then about 1.5 V. The temperature coefficient may be a problem - but it is worth an investigation.

 

[Skogsgurra]

Correction (green 3mm LED). Found my note-book.

The voltage drop is 1.70 V at 10 uA and 1.76 V at 40 uA and +20 C. These voltages drop to 1.66 and 1.72 V at about +70 C. No noticeable increase in leakage current when increasing temperature from 20 to 70 C.

The spread between LEDs is neglible - the colour determines energy levels and they determine voltage drop. So it should be quite constant.

 

[MacGyverS2000]

log, I realized my number foulup late last night, so yes, I'll go with 1%... this is a consulting project on the side of my real work (non-critical work, thank God), so I'm having a few too many all-nighters. The supply will be 9V, but batteries and power supplies being what they are, I'm giving myself a 10V headroom to avoid going above the max spec on the A/D channels. Divided down, that gives me a nominal 4.5V out of 5V to sample, which is fine.

Yageo offers 4.99k, 1% 0603 at less than half a penny a piece, and Susumo offers roughly the same at 0.5% if I really need it (I doubt it, the app isn't that critical).


skoggs, Useful idea, I'll keep that one in my back pocket.


scotty, your idea has merit, and I've considered something along those lines. Complexity has stopped me so far as I believe I can get the accuracy/repeatability I want with something more simple. I may change my mind in a day or two, but so far the resistor dividor with the FET is looking pretty good.

 

[IRstuff]

Have you accounted for the resistor temperature coefficients as well?

TTFN

 

[logbook]

IFstuff

1% resistors are typically 100ppm/degC or better, so over 15degC the resulting error is negligible compared to the 1% spec. If he uses a pair of equal valued resistors then one would have to suppose that their TCs will track anyway, giving another factor of say 5x improvement.

 

[MacGyverS2000]

+/- 100ppm for the 1% Yageos, and a nice 50ppm for the 0.5% Susumus

 

[nbucska]

Moc:
#1 doesn't work-- you can't use one single serial R for
the current is changing too much with temp. and chip to
chip. Voltage divider must use multiple of the chip's
current unless you buffer it with a voltage follower.

The resistor tolerance can be compensated by using a
cal. factor in the program, the termal variation can't...


<nbucska@pcperipherals DOT com> subj: eng-tips
read FAQ240-1032