Online Antenna Mode Selection for Reconfigurable MIMO via Adaptive Bandit

Reconfigurable antennas (RAs) can enhance link capacity by swiftly changing their radiation pattern to adapt to the wireless propagation environment actively. In order to exploit the property of RAs, selecting the time-varying optimal mode from candidate radiation patterns for MIMO communication is essential but however challenging in a mobile environment. In this paper, we focus on the efficient radiation mode selection for reconfigurable MIMO systems, where the transmit array consists of RAs instead of conventional antennas with fixed radiation mode. Channel prediction based on angle information is studied to obtain the channel state information (CSI) of unselected modes based on the estimated path gains. We further propose two bandit algorithms: upper confidence bound based on adaptive bandit with sliding window (UCB-AB) and Thompson sampling based on adaptive bandit with sliding window (TS-AB). UCB-AB and TS-AB exploit the useful parts of outdated CSI outside the sliding window and improve the convergence rate of bandit algorithms under mobile scenarios. Numerical results demonstrate that the introduction of adaptive bandit can enhance the convergence of bandit algorithms and improve the throughput of MIMO systems.