We use 0.1% resistors, and sometimes 0.05%. But they are still nailed
to the 1% values, which are geometrically spaced. So if you have some
assortment of resistors in stock, various combinations tend to keep
hitting the same ratios, and some ratios can't be done.
John
Ah, but for small quantities, it's much cheaper to buy standard
values. You _can_ get any value you want, if you pay enough; but if
your volume isn't great enough, it's much more economical to do series
and/or parallel combinations. With an E24 set of high accuracy
values, two in parallel or two in series will always get you within
0.23% of a desired value, and if you're stuck trying to get to a low
value where series won't work, or a high value where parallel won't
work, you still can get within 0.45%. And that's without being very
creative about picking values (just one closest to the desired and one
to trim it). For ratiometric stuff, it generally gets even better:
with one "stock" value and one trimmed value, you get to 0.23%, but
add a trim to the stock value and you get much finer ratio adjustment.
I have some very low TC matched resistor sets in my junque box, from
HP voltmeter scrap, that have "weird" values. Given the needed
precision and the relatively high volume in the voltmeter market, the
custom values made sense for HP. For the work I do, trimming as
needed with series and/or parallel is fine: I do need 0.1% parts at
times, but never in high volume.
Also, it helps a lot to design systems in such a way that really close
arbitrary ratios aren't necessary. For practically everything we do,
stability is far more important than absolute accuracy; the accuracy
is obtained by calibration. Calibration used to involve adjusting
pots and possibly variable capacitances and inductances; that's
practically all been replaced with processor-based calibration. I
realize that on rare occasion, there's no good way around some
precision ratio, but at least in what I've seen for quite a few years
now, those occasions are pretty rare.
Cheers,
Tom