Opportunistic medium access in multi-channel wireless systems: A learning approach

Abstract

Dynamic channel selection is an important component of multi-channel wireless systems. It allows a transmitter to identify the channel offering the best radio conditions and to avoid interference created by other transmitters. In absence of interference, the channel selection problem can be simply interpreted as a Multi-Armed Bandit problem for which low-regret learning algorithms such as Exp3 have been developed. With interference, the problem is complicated by the fact that transmitters interact, and that a given transmitter experiencing a transmission failure cannot identify whether the failure is due to a channel error or to interference. In this paper, we analyze the dynamics of a system where transmitters independently run the learning algorithm Exp3 to select channels for their successive transmissions. We show that the system converges to a pure Nash Equilibrium of the corresponding game.