2020/10/22 by Ryan M. Dreifuerst, Dreifuerst, Ryan M., Samuel Daulton +17 · 3 citations
Engineering · Computer Science · #Advanced MIMO Systems Optimization #Advanced Wireless Network Optimization #Wireless Communication Networks Research
paper · pdf · doi:10.48550/arxiv.2010.13710
Wireless cellular networks have many parameters that are normally tuned upon\ndeployment and re-tuned as the network changes. Many operational parameters\naffect reference signal received power (RSRP), reference signal received\nquality (RSRQ), signal-to-interference-plus-noise-ratio (SINR), and,\nultimately, throughput. In this paper, we develop and compare two approaches\nfor maximizing coverage and minimizing interference by jointly optimizing the\ntransmit power and downtilt (elevation tilt) settings across sectors. To\nevaluate different parameter configurations offline, we construct a realistic\nsimulation model that captures geographic correlations. Using this model, we\nevaluate two optimization methods: deep deterministic policy gradient (DDPG), a\nreinforcement learning (RL) algorithm, and multi-objective Bayesian\noptimization (BO). Our simulations show that both approaches significantly\noutperform random search and converge to comparable Pareto frontiers, but that\nBO converges with two orders of magnitude fewer evaluations than DDPG. Our\nresults suggest that data-driven techniques can effectively self-optimize\ncoverage and capacity in cellular networks.\n