Barry Jones said:
You really have to have *something* you trust to start with. Two
somethings makes it even easier.
I was thinking about finding *something*, and it seems to me, at least
theoretically, that you could make a large set of resistors (or other
things) of a given precision into a smaller set of resistors (or other
things) with higher precision.
For example, suppose you have 100 1 M resistors with a precision of 1%.
If you connect them all in parallel, you'd have an equivalent 10K
resistor, but it's standard deviation will have decreased by a factor of
sqrt(100). This is from the definition of Sample Normal Distribution.
Assuming the 1 M resistors had a mean of 1 M, and a somewhat normal
error distribution (it doesn't even have to be very close to normal),
this should increase the precision by a factor of 10. What say you to
the analysis?
Of course there may be other errors introduced in trying to connect 100
resistors in parallel. Details, details.
Like getting the bus hot enough to solder all 100 of them, which then
heats the other resistors up for a long time. Not so good.
But what if the testing machinery happens to be off a half percent that
day, on the high side, and all of the resistors in your batch are ones
made that day. Now your statistical curve is distorted significantly.
Now if I could just invent perpetual motion . . .
Maybe it's better to just scrounge a known good 0.1% resistor from a
piece of equipment, and have it checked by a known accurate meter. Sort
of like those sets of weights with the little ivory tweezers to pick
them up. They are known standards, cal'd at a lab that can trace back
to the standards at NIST or wherever. You don't use them day-to-day,
just once in awhile to verify that your instruments are working
properly.
