Pure-exploration bandits for channel selection in mission-critical wireless communications

In emergency communications, guaranteeing ultrareliable and low-latency communication is challenging yet crucial to save human lives and to coordinate the operations of first responders. To address this problem, we introduce a general approach for channel selection in mission-critical communications, i.e., choose channels with the best quality timely and accurately via channel probing. Since the channel conditions are dynamic and initially unknown to wireless users, choosing channels with the best conditions is nontrivial. Thus, we adopt online learning methods to let users probe channels and predict the channel conditions by a restricted time interval of observation. We formulate this problem as an emerging branch of the classic multiarmed bandit (MAB) problem, namely the pure-exploration bandit problem, to achieve a tradeoff between sampling time/resource budget and the channel selection accuracy (i.e., the probability of selecting optimal channels). The goal of the learning process is to choose the 'optimal subset' of channels after a limited time period of channel probing. We propose and evaluate one learning policy for the single-user case and three learning policies for the distributed multiuser cases. We take communication costs and interference costs into account, and analyze the tradeoff between these costs and the accuracy of channel selection. Extensive simulations are conducted and the results show that the proposed algorithms can achieve considerably higher channel selection accuracy than previous exploration bandit approaches and classic MAB methods.