After reading an example:
The bit period in the baseband signal is T = 1/200000 seconds, and the
baseband pulses are rectangular. The bandwidth of the baseband signal (to the
first null) is given by B = 1/(2T ). For the RF components, 2 d f = 150 kHz.
The bandwidth of the FSK signal is therefore
2 d f + 2B = 150 kHz + 200 kHz = 350 kHz.
I find it a little bit confusing, why author using coefficient 2 referring to d f, when it's just "spacing" between marks "0" and "1" frequencies?
This is a manipulation, and if I need to make things more difficult to understand, I'd exploit this technics .
Briskly searching on my e-book shelf, I find this explanation more clear:
When only a short time duration dT is available, the uncertainty theorem puts a constraint on the accuracy of recognizing the difference between the two frequencies df.
dF x dT >= 2 x PI();
For example, the telephone channel is less than 4 KHz wide, and so a rather large separation would be df x 2 KHz. This implies that telephone line FSK information transfer rates will not exceed around 300 b/s.
Digital Signal Processing
A Computer SciencePerspective
Jonathan (Y) Stein
Never heard of "Carsons rule". I like http://en.wikipedia.org/wiki/Uncertainty_principle