Dynamic Spectrum Access Using Stochastic Multi-User Bandits

A stochastic multi-user multi-armed bandit framework is used to develop algorithms for uncoordinated spectrum access. In contrast prior work, it assumed that rewards can be non-zero even under collisions, thus allowing the number of users greater than channels. The proposed algorithm consists an estimation phase and allocation phase. It shown if every user adopts algorithm, system wide regret order-optimal order O(log T) over a time-horizon duration T. guarantees hold both cases where or less extended dynamic case in evolves time, lead sub-linear regret.