To chime in late: typically the routine in LNA design is noise match
at the input, then conjugate match at the output considering the
effects of the noise match. Really the only reason to play games with
the input noise match is either for broadbanding or to address a
specific stability issue, which usually is better handled by judicious
feedback.


Joel Kolstad wrote:
Thanks again Tom, you've really cleared things up for me here.

Here's the concrete numbers... in the center of the band, S22 -- converted
back to an impedance -- is 119.5-j90 ohms. With a 2V source, if I conjugate
match my 50 ohm load obviously I get 1V across the resistor or 20mW = 13dBm.
Without a match... let's see... current around the loop is 2/(119.5-j90+50) =
9.2+j4.89 mA, and power in the 50 ohm load is I^2*R = 3.04+j4.5mW = (looking
just at real power) 4.83dBm. So... without a match... I'd be leaving 13-4.83
= 8.18dB of gain "on the floor"... ouch!

---Joel