The old theory, before Shannon showed us a better way, was 1 bit per second required 1 hertz. Under the newer theory, if you think about signal to noise power, then by using multiple levels of encoding larger than the noise power, you can exceed the 1 hertz per bit concept. Thus you can trade off bandwidth versus signal to noise signaling power with proper encoding and use greater transmitting power. In order to reach the theoretical improved data rates to bandwidth, avoidance of differential phase and gain errors in the transmission is achieved by predistortion prior to transmission, which will smear the data phase and amplitude constellations. Thus the maximum data rate for a channel is set by the expected ambient noise power, how much you have in your power budget to send the information, and an acceptable bit to error ratio. An example is MELPE vocoder transmissions using 200 Hz bandwidth that have measured as providing greater intelligibility at the far end than an unencoded transmission.