[sdiy] phantom power: how to avoid electrolytics?

[jhaible]

>I doubt the BC549 will be close to a ultra matched pair in their noise
>specs. (This is an app that is demanding the large geometry devices...)

Doesn't this National Semiconductor low noise pair have something
like 100 transistor pairs inside? (forgot the part number. LM194 ?)

If you want to do this with discrete transistors, questions arise.

In theory, the signal adds arithmetically (correlated), and the
noise adds geometrically (not correlated; power adds instead of voltage
or current.) So connecting a lot of transistors in parallel will
increase SNR. If all transistors are equally noisy, N transitors will
produce N times the signal, and SQRT(N) times the noise.
(Is this formula right?)
So your SNR would be increased by 20log(N/SQRT(N)) dB =
20 log SQRT(N) dB.
Two pairs will give 3dB (the same amount you loose if you're
using a pair instead of a single transistor, btw!).
100 pairs will give 20dB.
More realistic for diy:
10 pairs will give 10dB over 1 pair.
(Think of it: 20 transistors connected as 10 pairs will only give 7dB
improovement over a single transistor stage!)

But I wonder what happens if the transitors are *not* equal.
(Just speaking of noise here.)

Just imagine one of your 20 transistors (10 pairs) is considerably
more noisy than the others. Then your 10 pairs in parallel might
be more noisy than a single, selected pair. It might be more noisy
than a randmoly chosen pair, if you are lucky.

Does this make sense? I guess it depends on how big the variance of
noise between specimens of a specific transitor type actually is.
(If the variance is small, paralleling will always give an improovement.)

JH.