Reinforcement learning system to mitigate small-cell interference through directionality

Beam-steering techniques using directional antennas are expected to play an important role in wireless network capacity expansion through ubiquitous small-cell deployment. However, integrating directional antennas into the existing wireless PHY and MAC stack of small cells has been challenging due to the added protocol overhead and lack of a robust antenna beam selection technique that can adapt well to environmental changes. This paper presents the design, implementation, and evaluation of LinkPursuit, a novel learning protocol for distributed antenna state selection in directional small-cell networks. LinkPursuit relies on reconfigurable antennas and a synchronous TimeDivision Multiple Access (TDMA) MAC to achieve simultaneous directional transmission and reception. Further, the system employs a practical antenna selection protocol based on the well known adaptive pursuit algorithm from the reinforcement learning literature. We implement a realtime prototype of LinkPursuit on the WARP platform and conduct extensive experiments to evaluate its performance. The empirical results show that appropriate use of directionality in LinkPursuit can result in higher network sum rates than omnidirectional transmission under various degrees of cross-link interference.