Efficient Fair Probabilistic Multi-armed Bandit for Real-Time Resource Allocation in Spectrum Sharing

Spectrum sharing is critical to addressing spectrum shortages driven by the rapid growth in mobile users and data traffic. Interference mitigation poses a significant challenge to realize efficient and fair spectrum resource allocation in dynamic environments. In this paper, we introduce an efficient fair probabilistic multi-armed bandit (EFP-MAB) algorithm, which optimizes the concurrent link transmission schedules to enhance coexisting networks’ throughput, by exploiting interference cancellation techniques. We provide a comprehensive performance analysis of EFP-MAB, deriving a bound on its expected regret. The expected regret of EFP-MAB improves by $O(\sqrt {K\log (T)} )$ compared to the state-of-the-art FP-OFU algorithm, where T is the time horizon and K is the number of meta-arms. The computational complexity of EFP-MAB is poly(N,K), where N represents the number of arms, marking a significant efficiency improvement over the FP-OFU algorithm, which lacks polynomial time global optimality guarantees. Extensive simulations in a heterogeneous technology coexistence scenario demonstrate that EFP-MAB significantly outperforms FP-OFU in both expected regret and efficiency.