I understand noise computations. [In fact, my comment on it being
proportional to the square root of bandwidth is because you need to
multiply the input referred noise in v/root hertz by the square root of
the bandwidth, assuming the noise is uniform. Conversely, take the
square root of your equation.] You have mixed in way too many side
issues. So are you saying by 18db SNR, he meant a noise figure of 18db?

I've never seen the measurement done, but I assume when a radio specs a
sensitivity at a given SINAD, there is a distortion analyser on the
audio output. The filtered signal is the "S", and the "NAD" is the
residual of the signal.