I just saw this thread and noted, by the way, prompt accurate information
from IsaacWingfield responding to the original question.

"Detector195" in om...

. . . I took two resistors as
"signal sources" to compare their noise spectra.
One was a 100k 1/4 W metal film, the other was
a 100k 1/8 W carbon film.

(Careful -- two variables changed at once. An issue of possible distraction
to some readers, more really than of misleading conclusions.)


The carbon film had 6 percent more voltage noise from essentially
zero up to 20 kHz, give or take 1 percent. This is about 0.5 dB.
Does anybody out there have a real "professional" spectrum
analyzer or digital scope? This should be an easy measurement
if you have the right equipment.

Those data sound reasonable to me. Without going into details, I have done
many measurements as queried above. I use some very exquisite equipment for
low-noise measurements, some of the best available in the last few decades;
we have a lab full of it for such purposes. (By the way, you generally want
analog, or at least good-analog-front-end, instruments to measure low-freq.
noise spectra with low floors. And/or, build outboard low-noise preamps for
the purpose and keep them in the same lab.) My end applications were not
specifically audio, but I measured in the same frequency range. (I was
concerned with "making" resistors in monolithic form, and possible sources
of "excess" noise).

The whole issue here is the "excess" noise that Isaac Wingfield already
concisely explained. Typically it arises not in uniform resistive materials
but in interfaces between different materials, or between different crystals
or clusters of material. Commercial component resistors of metal or metal
film, common and fairly cheap anyway, have the general reputation of the
lowest excess noise, followed by carbon film. Carbon composition resistors
(polycrystalline I think, and very common construction for wired reasistors
when I was younger) have the worst reputation.

Issues of the power rating of a resistor are only peripherally related to
its noise sources. Self-heating capable of significantly raising a
resistor's Kelvin temperature (which is what counts) from the usual 300-350
found in operating electronic equipment would mean that the resistor is run
at a high power level, which argues independently for a higher power-capable
resistor, regardless of noise considerations. (Resistors, like people, are
most reliable when not overheated.)

I still think Isaac Wingfield said it more concisely. -- Max Hauser