Unsupervised classification of non-linear dynamics in optical fiber propagation using intensity clustering

We demonstrate that centroid-based clustering of normalized intensity profiles is able to successfully isolate different classes of pulses associated with physically distinct regimes of nonlinear and dispersive propagation in optical fiber. Remarkable for its simplicity, this approach shows how analysis of only the temporal intensity profiles of propagating pulses, even at relatively limited sampling resolution, reveal sufficient similarities to allow physical classification of different classes of propagation behavior.