Distributed Algorithm for Opportunistic Spectrum Access in Dynamic Ad Hoc Networks

The opportunistic spectrum access (OSA) algorithms allow secondary users (SUs) to exploit vacant channels with an aim to maximize overall spectrum utilization/throughput. The design of OSA algorithm is challenging for ad hoc networks due to lack of coordination among SUs and unknown channel statistics. It becomes even more challenging for the dynamic networks where the SUs can enter or leave the network any time without prior agreement. Most of the existing algorithms assume either prior knowledge of the number of SUs or need wideband sensing to sense all channels simultaneously to guarantee optimal channel allocation among SUs. Our goal in this paper is to develop distributed OSA algorithm for dynamic ad hoc networks that offers higher throughput without compromising on the number of SUs collisions. The proposed distributed algorithm is based on multi-player multi-arm bandit framework and they allow SUs to independently estimate the number of other SUs and channel statistics. We derive the upper bounds on the throughput loss (regret) and number of collisions. Exhaustive synthetic results and experimental results on universal software radio peripherals (USRP) based testbed validate our claims and superiority of the proposed algorithm.