Thanks, but I've been woefully inadequate at getting my point across.
The methods suggested are charge balance measurements.
A sigma-delta converter is a big bucket. There's a hole in the bottom and the size of the hole represents the quantity being measured. You sit at the top and teaspoon more water in to keep the water level constant. The number of teaspoons per unit time has to equal the volume of water running out the bottom.
Similarly, a single slope converter lets the water level fall to a threshold and you dump a whole lot of water in the top to bring the level instantly back to the top threshold. Same concept, bigger spoon and you only do it once/cycle. Time measurement vs count... potato...potato...
These are both current controlled devices. You're measuring the current produced thru the resistor from the input to the reference voltage for sigma-delta and the current charging the cap in the slope converter.
Neither of these are very helpful when you want your input voltage dynamic range to be half the offset spec of the comparator.
In the sigma-delta case, it's like trying to keep the level in a
5-gallon bucket constant by pouring from a peanut butter jar into a straw sealed into the top of the bucket without a funnel. Some of the water gets in, some doesn't. The "slop" in the system exceeds the quantities being measured.What I'm looking for is a CLEVER topology that gets rid of the slop. It's the funnel in the above analogy. You're more likely to find such an item in a patent disclosure than in "A/D for dummies".
The obvious "funnel" is a voltage to current converter that gets the signal into the proper paradigm for these types of converters. Might even be able to do it with a modified current mirror. The nice thing about current is that the voltage on the cap keeps changing until it reaches the threshold...even if the threshold ain't exactly where you expect it. You can calibrate out systematic errors. You can't do nothin' if the voltage never gets to the threshold.
Somebody asked for numbers. The numbers I gave in the first post are what I want. 10mV full scale and 1% resolution. But I did dig out a real voltmeter. I wrapped the peltier device in cellulose...aka a paper towel, stuck it into the plastic lid of a spray can and set it on the desk. After a LONG time, the voltage did settle to about 100nanovolts. That's the best this old meter can do. I was surprised it did anywhere that well as I took zero precautions to balance the thermals in the interconnect. I can see significant voltage transient if I flex the coax that connects it.
If I take it out of its cocoon, the voltage is all over the place. Stick it on a window with 14F inside/outside temperature differential, voltage goes to 100mV or so and dribbles down to about 2mV. It's the
2mV I want to measure. my 1% resolution is worth 5% at 20% of FS.All seems consistent with my initial estimates.
The back story is this... I got a new gas furnace and a bunch of insulation added to the house. I've been graphing heat input. Consistent with my minimalist approach, I used a flapper on a microswitch sitting on a vent and plugged into the serial port of a Palm IIIc...and fixed it in software.
I've been seeing significant anomalies that I can't explain by outside air temperature fluctuations. Consistent with my lazy approach, the measurement system was a bottom-up design...who am I kidding, it wasn't designed, it was evolved... I had no idea what I was measuring. After I did the math, I discovered that the system has a RESOLUTION of 17 BTU. I don't claim that it's anywhere near that accurate, but it does resolve well. Turns out that I can see when the computer goes into standby or when I'm watching TV mirrored in gas consumption.
By this time, I'm into measuring losses. My windows are specified as R3.125. But I didn't measure anywhere near that high. And when I added R5 foam on the outside the improvement was more than I expected from R8.125/R3.125. That's where all the peltier device stuff started.
So, I have no idea how they measure the insulating capabilities of windows, but the numbers seem optimistic. This is an interesting reference:
It's the operator/application manual for their IR thermometer tweaked to measure building heat loss.
If you sandwich the window between two metal plates and measure R, you get a small number. If you measure ambient air temperature in still air on both sides, you get a much bigger number. If you use the method outlined in the link, you get a third number.
There are all kinds of issues with boundary layers, convection, wind etc. on both sides of a window. I started looking into those effects. And I need some repeatable method to measure it.
Why do I want to do this? Because I can...and I'm bored.
Still looking for that clever way to resolve small voltages with no hardware.