Online Learning for Spectrum Sensing and Reconfigurable Antenna Control

Efficient dynamic spectrum access (DSA) policies rely on accurate spectrum sensing information to exploit spectrum white space optimally. Obtaining accurate channel state information (CSI) from local spectrum measurements is made difficult by wireless fading and the presence of thermal noise which distorts measured signals and leads to uncertainty regarding the occupancy of spectrum resources. Electrically reconfigurable antenna systems (ERAS) offer the system designer an additional degree of freedom to exploit pattern and polarization diversity to improve the accuracy of local spectrum sensing decisions. We propose a learning technique to exploit pattern and polarization diversity offered by ERAS to improve spectrum sensing accuracy. While the proposed approach is designed to work within the unique design constraints of reconfigurable antennas, the approach is not antenna specific and will work with a wide variety of reconfigurable antenna designs. Validation of system performance is provided from measurements taken using the wireless open access research platform (WARP) software defined radio (SDR) platform in an indoor office environment.