Faster-Than-Nyquist Signaling

The degradation suffered when pulses satisfying the Nyguist criterion are used to transmit binary data in noise at supraconventional rates is studied. Optimum processing of the received waveforms is assumed, and attention is focused on the minimum distance between signal points as a performance criterion. An upper bound on this distance is given as a function of signaling speed. In particular, the pulse energy seems to be the minimum distance up to rates of transmission 25 percent faster than the Nyguist rate, but not beyond. Some mathematical aspects related to the above problem are also considered. In particular, the minimum distance is rigorously shown to be nonzero for all transmission rates. This is tantamount to showing that, in the singular case of linear prediction, perfect prediction cannot be approached with bounded prediction coefficients.