The real challenge here is to translate it into a musical form that anybody would actually listen to. There are (nowadays lots of) tools that make graphics out of sound (like CTHUGHA), and there it also depends on the parameters.

Why not add some pseudo-random stuff or even some melodic or harmonic stuff from a database, which has to be discarded in case someone wants to decode the original information, but which would make the whole track sound better. Like in J.S. Bach's time there was a vivid scene of musical improvisation, they just noted noted some single notes or numbers which were then performed in a more elaborate way, quite similar to modern guitar notations in songbooks. But then it must be done in a way that the 'naked' melody (i.e. information) is still recognizable. I'm not a musician, and no radio specialist either. Is it really possible? If not, why not?