2019/03/15 by Eren Balevi, Jeffrey G. Andrews, Balevi, Eren +1
Engineering · #Advanced MIMO Systems Optimization #Energy Harvesting in Wireless Networks #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Millimeter-Wave Propagation and Modeling
paper · pdf · doi:10.48550/arxiv.1903.06787
openalex publication_date 2019/03/15 · openalex created_date 2022/07/29 · openalex updated_date 2026/07/28
We aim to jointly optimize antenna tilt angle, and vertical and horizontal\nhalf-power beamwidths of the macrocells in a heterogeneous cellular network\n(HetNet). The interactions between the cells, most notably due to their coupled\ninterference render this optimization prohibitively complex. Utilizing a single\nagent reinforcement learning (RL) algorithm for this optimization becomes quite\nsuboptimum despite its scalability, whereas multi-agent RL algorithms yield\nbetter solutions at the expense of scalability. Hence, we propose a compromise\nalgorithm between these two. Specifically, a multi-agent mean field RL\nalgorithm is first utilized in the offline phase so as to transfer information\nas features for the second (online) phase single agent RL algorithm, which\nemploys a deep neural network to learn users locations. This two-step approach\nis a practical solution for real deployments, which should automatically adapt\nto environmental changes in the network. Our results illustrate that the\nproposed algorithm approaches the performance of the multi-agent RL, which\nrequires millions of trials, with hundreds of online trials, assuming\nrelatively low environmental dynamics, and performs much better than a single\nagent RL. Furthermore, the proposed algorithm is compact and implementable, and\nempirically appears to provide a performance guarantee regardless of the amount\nof environmental dynamics.\n