Opportunistic Spectrum Access via Good Arm Identification

Zhiyang Wang, Ziyu Ying, Cong Shen

Abstract: In this work, we promote a different tool of multi-armed bandits (MAB), called arm identification, to choose a suitable channel for Opportunistic Spectrum Access (OSA) with proven accuracy while satisfying stringent constraints on delay, energy consumption, and channel switches. Noting that finding the best channel may not always be the optimal choice, we deviate from the celebrated best arm identification framework and adopt good arm identification (GAI), which results in a channel that is "good enough", but requires much less time and energy consumption under the same accuracy requirement. Robustness issues such as delayed or missing feedback are also studied under the new framework. Performance of the proposed algorithm is studied analytically and further corroborated via numerical simulations.