Reduction of collisions and regret in time sharing schemes for opportunistic spectrum access

We examine decentralized learning and access algorithms for opportunistic spectrum access with multiple users. Several distributed algorithms have been proposed for this problem, mainly as an application of corresponding algorithms for the multiarmed bandit problem, which are provably order optimal in terms of regret. However, none of them pays particular attention to reducing collisions among users caused by lack of message exchanges. The effect of such collisions becomes more observable as the number of users increases, causing a considerable amount of added regret, despite retaining the optimal order. Focusing on time division fair sharing schemes based on the idea of orthogonal offsets, we propose a simple algorithm for detecting offset collisions and trying to resolve them as quickly as possible, inspired from persistent distributed schemes for multiple access. We demonstrate the improved performance achieved by our algorithm by means of simulations.