http://www.conknet.com/~b_mobile/NoiseStuff.html
This link should clarify things.

In summary, the theoretical value of noise is: [E^2 - tRkTB] where :

E is the RMS value of noise,
R is in ohms
T is in absolute temperature,
B is the bandwidth in Hz
k is Boltzmans constant (k = 1.38 x 10^-23 joules/degree K)

At room temperature (290 degrees K) with a 50 ohm system with a 1kHz
bandwidth, this converts to a noise level
of -144dBm. 15dB excess noise from the environment converts to a level of
(-144 + 15) or -129dBm. This translates to a level of slightly less than
0.1uV. This is the noise level that you might find in a very quiet rural
area. In my area just north of Chicago, the excess noise is around 4uV at
30MHz, increasing to higher levels on the lower frequencies. At frequencies
below 100MHz, the predominant noise is man-made noise and atmospheric noise.
Above 100MHz, the noise density decreases sharply with galactic noise being
the dominating factor. This is the reason that a receiver with a 12dB NF is
acceptable for reception below 30MHz. When you listen to receivers that seem
to be quieter than others on the same signal, this can be because of noise
in the audio amplifier or because of excess phase noise in the LO
(synthesizer). Excess phase noise manifests itself in a couple of ways;
first of all, it can cause that "white noise" that is sometimes experiences
on signals, but more often it is because of signal feedaround in the I.F.
filter chain. The second way that excess phase noise manifests itself is by
the rising of the system noise floor when tuning very close in frequency to
a very strong adjacent signal.
Two ways that I know of to clean up this problem a use a selective
crystal filter (500Hz BW) at the Phase Detector's reference oscillator input
in the synthesizer. The second way (mainly to improve close-in IP3) is to
use a multi-pole filter with good shape factor as the roofing filter. The
only problem here is when you use a multi-loop synthesizer to tune the minor
loop at the 2nd LO frequency. The Racal 6790/GM addresses this problem by
using a single loop Fractional N synthesizer. That is why they can get away
with using that (8 or 10-poles?) filter at the 45MHz 1st I.F.
As far as receiving higher frequencies, the benefit of a system NF of 2 to
3dB becomes apparent. There are GaAsFET devices that have a NF of a fraction
of a dB,but those are more useful for EME communications, etc.
Anyway, I hope this clarifies things.

Pete

..
..message oups.com...
I'm really kind of baffled what the guy means my 18 db signal to noise.
It's kind of meaningless without knowing how it is measured. Noise is
propertional to the square root of the bandwidth, so you need to know
what filter is being used.

I'm not sure you would use a noise figure meter. Seems to me you would
want the SNR measured from RF input to speaker output because that is
how you use the radio.

I may do this measure some day for yucks, but it will be a bit of work.
I'd have to fire up the old HP334A just to measure the THD of my RF
generator, not that I ever used the box in that mode.