Optimizing Coverage and Capacity in Cellular Networks using Machine Learning

Abstract

Wireless cellular networks have many parameters that are normally tuned upon deployment and re-tuned as the network changes. Many operational parameters affect reference signal received power (RSRP), reference signal received quality (RSRQ), signal-to-interference-plus-noise-ratio (SINR), and, ultimately, throughput. In this paper, we develop and compare two approaches for maximizing coverage and minimizing interference by jointly optimizing the transmit power and downtilt (elevation tilt) settings across sectors. To evaluate different parameter configurations offline, we construct a realistic simulation model that captures geographic correlations. Using this model, we evaluate two optimization methods: deep deterministic policy gradient (DDPG), a reinforcement learning (RL) algorithm, and multiobjective Bayesian optimization (BO). Our simulations show that both approaches significantly outperform random search and converge to comparable Pareto frontiers, but that BO converges with two orders of magnitude fewer evaluations than DDPG. Our results suggest that data-driven techniques can effectively self-optimize coverage and capacity in cellular networks.

Bayesian Optimization for Radio Resource Management: Open Loop Power Control

“…5b). To achieve this, we used a covariance Matérn kernel function 𝐾 𝑀 𝜃 defined as in (7). We simply initialized the hyperparameters ℓ so as to homogenize the scale of the two variables of x = [𝛼, 𝑃 0 ], i.e., ℓ 𝛼 /ℓ 𝑃 0 = 226.…”

“…In our experience, Bayesian optimisation with Gaussian processes (BOGP) 1 is a very powerful tool, which is well adapted to a wide range of RRM parameter tuning tasks, and that addresses the above challenges to a large extent. As shown recently in [7] for the task of transmit power and antenna tilt optimization, BOGP can achieve convergence within two orders of magnitude fewer iterations than a modern RL-based solution. A remarkable additional advantage of BOGP is that it allows to naturally embed prior knowledge from experts, simulations, or real-world deployments in the optimization process.…”

“…In practical RRM applications, one should choose: i) the kernel function that is believed to suit best the shape of the system performance, and a reasonable initial value for both ii) the kernel hyperparameters 𝜃 and iii) the observation noise variance 𝜎 2 . Concerning i), we believe that the Matérn kernel (7) is sufficiently flexible for a large range of functions. Items ii) and iii) are greatly facilitated by the possibility of tuning the hyperparameters in real-time to make the covariance function "adhere" at best to the observations in a maximum-likelihood sense, as described in Section III-D.…”

“…Prior 3 is the mirror of Prior 0. All priors share the same covariance matrix, which is a Matérn kernel function 𝐾 𝑀 𝜃 as defined in(7).…”

Cellular Network Capacity and Coverage Enhancement with MDT Data and Deep Reinforcement Learning

“…One way to tackle CCO is to allow coverage regions to change by modifying cells' elevation tilts. Some early decade contributions made use of research operations methods [12,13] to tune antennas parameters, while most recent years, motivated by the late success of Deep Learning (DL) and Reinforcement Learning (RL), have seen the rise of solutions employing DRL algorithms [14][15][16][17]. RL, indeed, seems to be a valuable tool for CCO since it can learn and adapt to the dynamics of the environment [14].…”