Online learning in decentralized multi-user spectrum access with synchronized explorations

In this paper we consider decentralized multi-user online learning of unused spectrum bands as an opportunistic spectrum access (OSA) problem. There is a set of M secondary users exploiting the spectrum opportunities in K channels. We develop a distributed algorithm for the secondary users that will learn the optimal allocation with logarithmic regret. Thus, our algorithm achieves the fastest convergence rate to the optimal allocation. In a more general framework, our algorithm gives an order optimal solution to the decentralized multi-player multi-armed bandit problem with general reward functions.