I think for now I'll live without the reference, and just make sure I don't let the battery get too drained.

Another idea I had is to use a voltage divider to measure the battery voltage directly and display a low-battery warning before it gets low enough to sag Vdd. That doesn't solve the Pickit 2 issue (not that it's a major issue, just for testing). It also means tying up another I/O pin which I can't spare, or switching to a PIC with more pins.

Dang... I just thought of something... my whole issue is that the MCP9700 sensor has an internal reference voltage so that it always outputs X volts for Y degrees of temperature. Of course, I can't access that reference, so I need another reference so that my readings are accurate.

But... someone said that silicon diodes have a pretty linear temperature response in its voltage drop. I could use that as my temp sensor, foregoing the 9700 entirely, and then by using Vdd as my A/D reference and the source power for the diode, the circuit would be self-referencing. Hmm... I might have to experiment with that.

Diodes do not have linear V-I curves, so they will not linearly track the bias voltage source (for self-referencing). They have a TC of ≈-2mV/deg C, but their forward voltage and exact TC varies from type to type and unit to unit within a type, and also with current. This means you would need to calibrate them with at least two temperature reference points.

Hi again,


Some of the other posts reminded me of another way :-)

Use a few diodes in series as a voltage reference (maybe >2.5v).
Use a thermistor and measure the temperature using the diode voltage reference
even though it will be off.
Using both measurements, calculate the real voltage of the diode and
include that in whatever actual measurements you have to make.

To calibrate this scheme, run the diode and thermistor at room temperature,
measure the voltage and the thermistor and record. Turn up the temperature
by 10 degrees C and measure both again and record. You may have to
do a third time if you need better accuracy.

What will happen is the diode voltage will change and so will the thermistor,
with the diode reacting to temperature mostly and the thermistor will vary
both because of temperature AND voltage so it should be possible to calculate
the true voltage on the diode knowing these two pieces of information because
for any given temperature the thermistor will still read the same resistance,
just a different divider voltage than when the source voltage is constant. In effect,
it will be like a thermistor squared :-)
If you need help with the equations just yell.

It is also possible with this scheme to allow the reference voltage to vary widely,
and measure the voltage of a single diode, then do the calculation for what
the reference voltage is based on the measured voltage of the diode and of
the thermistor (in a voltage divider circuit).

It was also common practice a long while back to temperature compensate a circuit
with a thermistor. In this way it may be possible to compensate the diode(s) with
a thermistor, but it's not that easy to do.

Originally Posted by kpatz

In poking around on the 'net I see some folks have used LEDs as zeners for voltage references. Any thoughts on this, or are they also too sensitive to temperature?

On p.3 of this article, the tempcos of various LEDs were measured:

IR........-2.3mV/deg C
Blue.....-5mV/deg C
Green...-3.8mV/deg C
Red......-1.5mV/deg C

Originally Posted by kpatz

(I'm not giving up easily...)

How 'bout this? A JFET current source, and tap a 2.5V VRef off the resistor as shown in the attached screenshot.

In the simulation, it is stable down to about a 3.1V supply voltage, which is more than sufficient. But is it (reasonably) thermally stable?

Add this to your sim:
.step temp 25 50 5
or whatever you want. This will run 6 sims, starting at 25C, ending at 50C, in 5C steps.
You can also do a DC sweep instead of a transient sim. Set the supply voltage to some arbitrary voltage, then sweep it in small steps. I did 0 to 5V in 1mV steps. You get a nice plot of Vout vs Vcc.

Even better than SPICE is an actual breadboard. With an actual MPF102 and using a 3.9k in parallel with a 180K resistor (giving me an actual value of 3.77k according to my multimeter) I get 2.500 volts on the dot.

As for temperature stability, I only saw a change of 1mV when I touched an ice cube to the top of the FET. I think it's good enough for my purposes.

EDIT: I spoke too soon. I had 2.5V at the output when I fed the circuit with 5V. When I feed it with 9V, I get 2.532V. Not exactly self regulating voltage wise, and not what I got with Spice...

Originally Posted by kpatz

Even better than SPICE is an actual breadboard. With an actual MPF102 and using a 3.9k in parallel with a 180K resistor (giving me an actual value of 3.77k according to my multimeter) I get 2.500 volts on the dot.

As for temperature stability, I only saw a change of 1mV when I touched an ice cube to the top of the FET. I think it's good enough for my purposes.

EDIT: I spoke too soon. I had 2.5V at the output when I fed the circuit with 5V. When I feed it with 9V, I get 2.532V. Not exactly self regulating voltage wise, and not what I got with Spice...

Why would you test over such a wide range? Can't you power it from your 78L05?

Originally Posted by kpatz

Well, if you built a voltage regulator or reference wouldn't you test it over a wider range than you'd see in the actual circuit? My simulations showed it to have a steady voltage over a wide input range.

Also, one of the purposes of this reference is to be able to detect a low battery by detecting sag on the 5V rail. So if I run this reference off the 78L05 and it sags as well, it won't work too well for that purpose. Otherwise, I'd just use a voltage divider off the 5V to get my reference.

But that is a good point... I should test it between 4.5-5V to see how much the output varies over that range. But, in theory at least, the output voltage shouldn't change, if you run a constant current through a fixed resistor, you should get a fixed voltage, right? (as long as the input voltage is high enough). Ohm's Law at work.

Yeah, that was my point. Test over the range that you expect to see.
A single transistor "constant" current source will never be absolutely constant as a function of Vcc.

How much current do you need? Will your PIC run at 3.6v? If you can get it down to about a 100uA, a Li-SOCl2 AA battery will last you 5 years. Their voltage output is very flat until they're dead (so little warning before they die) but they have a pretty good energy density and VERY low self discharge.

Do you really need much stability? If you want to know when the 5v line starts to sag, just measure the voltage across a diode as has already been brought up. There seems to be a lot of talk about how accurate it is and how stable over temperature it is, but it doesn't seem like any of that matters. You just need to know when the 5v starts to sag, and you would do so by seeing the apparent forward voltage across the diode rise. As long as its not linear and you chose a good threshold, I'd suspect you'd be ok. Is 2mV/Deg C (or whatever was stated) a big enough deal to make this a burden? I'm guessing this will be used where temperature can range 30 deg C over the course of a day (60mV swing, or 30mV around a center point). I'd just try a 100mV threshold.

Only problem is that once the 5v starts to sag, you probably don't have much time left anyway. Of course that all depends on the current you're drawing, and all the diode detection is going to do is waste additional power.

Noggin, you seem to be barking up the wrong tree. He needs a stable reference for his A/D, so he can accurately measure temperature with a temperature transducer which he already has.
If that's the issue you were addressing, it was not evident to me.