Coordinated beam selection in LTE-Advanced HetNets: A reinforcement learning approach

Endowing Heterogeneous Networks (HetNets) with Self-Organizing Network (SON) capabilities is instrumental in helping operators cope with small cell densification and network heterogeneity. In this paper, we assess how coordination among neighboring small cells helps control interference so as to improving the cell-edge throughput. The strategic coexistence between the macro- and small cell tier is examined, in which MIMO-capable small cells coordinate their downlink transmission, while adhering to the cross-tier interference constraint. Inspired from reinforcement learning (RL) framework, the proposed solution is totally decentralized, in which small cells jointly estimate their long-term utility functions, and optimize their beam selection strategies based on a precoding matrix indicator (PMI) feedback. The proposed beam selection procedure is validated in an extensive LTE-A HetNet simulator with a comparison with existing solutions, such as the Minimum Mean Square Error (MMSE) based beam selection method and uncoordinated MIMO transmission. Interestingly, the coordination-based solution is shown to yield substantial gains in terms of cell-edge throughput of up to 230% compared to the MMSE-based approach and 170% compared to the uncoordinated transmission.