Distributed opportunistic spectrum access with spatial reuse in cognitive radio networks

We formulate and study a multi-user multi-armed bandit (MAB) problem for opportunistic spectrum access (OSA) that exploits the temporal-spatial reuse of PU channels so that SUs who do not interfere with each other can make use of the same PU channel. We propose a three-stage distributed channel allocation policy for OSA, where SUs collaboratively find an optimal channel access grouping, and independently learn the channel availability statistics to maximize the total expected number of successful SU transmissions. We adopt a distributed synchronous greedy graph coloring algorithm to cluster SUs into maximal independent sets, and a distributed average consensus algorithm to learn the sizes of the independent sets, with SUs belonging to a larger set being assigned a smaller access rank. Each SU then independently learns the PU channel statistics using a revised ε-greedy policy based on its assigned access rank. We provide the theoretical upper bound for the regret, and simulations suggest that our proposed policy has a significantly smaller regret than a random access policy and an adaptive randomization policy.