Enhancing blind interference alignment with reinforcement learning

Abstract

Blind interference alignment (IA) is a signaling scheme that suppresses interference in multi-user systems, without the knowledge of channel state information at the transmitter (CSIT). The key to performing IA without CSIT is the use of reconfigurable antennas (RA) that are capable of dynamically switching among a fixed number of radiation patterns to introduce artificial fluctuations in the channel. The radiation patterns used to realize blind IA have significant impacts on the overall performance of the system. Hence, an intelligent antenna pattern selection strategy is a crucial component of any practical RA-based blind IA implementation. In this work, we propose two reinforcement learning algorithms for selecting the optimal antenna configuration for blind IA. Furthermore, we evaluate the performance of these antenna mode selection techniques using over the air measurements on our software defined radio implementation of blind IA using a Reconfigurable Alford Loop Antenna that is capable of generating multiple radiation patterns. We quantify the performance of the algorithms in terms of received signal to interference and noise ratio (SINR) and show that our learning-based mode selection strategies are capable of choosing the highest performing mode 90% of the time and attain over 2 dB gain in SINR over other selection approaches.