[sdiy] Resistor War !!!

[Martin Czech]

SNIP
:::So how can two different resistors of the same value produce different noise
:::figures? Are we talking about mechanical noise sources such as microphonics
:::or vibration sensitivity?
:::
:::Otherwise that does not make any sense. Resistors are not active voltage
:::sources, so how can one type produce more noise than another? (since they
:::don't actually generate additional noise at all?)
:::

any resistor, even ideal resistor will generate noise due to thermas
fluctuations of the electron cloud. The noise voltage is related to
resistance and temperature.

Un=SQRT{4*k*T*B*R}

is the effective (hissing white) noise voltage produced by an ideal resistor R
at temperature T (Kelvin!), k is Boltzman's constant and B bandwidth in Hz.
Example: you are interested in the effective noise voltage of a 10Meg
reisistor, and 20kHz bandwidth @ room temperature:

4*1.38E-23 Ws/K*300K*20000/s*10E6V/A=3.3E-9 V*V

Un=57.5uV

Quite a lot, eh? If your signal is 1Veff and this noise
volatge will simply add to it, this means only 84dB S/N.
This means only 14bit resolution, in the age of 16bit this
is no good.

They only way to avoid this is low impedance, or cooling.

Now, real resistors are not ideal. They have more noise, which 
is called excess noise. Carbon types are known to be more noisy in this
respect, because they are not made out of (metal)-crystalls, but carbon
chunks. You know the old carbon microphones? They sound pressure
compresses the carbon pieces, this will change resistance. In a 
carbon resistor thermal fluctuations and maybe vibrations can do
a similar effect, the resistance is modulated this genearting a noise
voltage if a current flows through the resistor. This excess noise
is not white, it follows the 1/f law. This means that the carbon particle
fluctuations are predominantly slow (which makes some sense).
Metal resistors have also 1/f noise, but the "corner" is much lower.
Carbons may have the 1/f corner right in the audio band. And for
precision dc work 1/f is a catastrophy, too.

RF circuits have other needs: inductance must be low. Carbon
type resistors may have an advantage with laser trimmed spiral wound
films, or even wire wounds. 1/f is of no importance there.

m.c.