Transitions from DCA to FCA behavior in a self-organizing cellular radio network

This paper studies self-organizing in a cellular radio network with dynamic channel assignment (DCA). Conceptually, channel assignment is split into two subsystems, one performing real-time assignment and one adjusting assignment probabilities, aiming to prefer selection of channels less affected by interference in the past. The latter subsystem is modeled by a differential equation. The resulting "learning" network is able to perform phase transitions from a fully dynamic state in which all channels are used with equal probability to a state in which each cell prefers channels which are less frequently chosen by interfering cells. Both theoretical considerations and numerical simulations show that the simple strategy to avoid interference is generally unable to provide optimum channel reuse distances. The situation can be improved by refined learning strategies. Further numerical simulation of inhomogeneous networks indicates that channel reuse performance achieved by this strategy is reasonable, but still inferior to that of dedicated channel assignment algorithms.