Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-driven Optimization and Transfer Learning for Cellular Network Antenna Configurations

Mohamed Benzaghta, Giovanni Geraci, David López-Pérez, Alvaro Valcarce

TL;DR

This work presents a data-driven, high-dimensional Bayesian optimization framework to jointly optimize base-station tilts and vertical HPBWs in a real London cellular network, using site-specific 3D ray-tracing with Sionna to model propagation. The approach employs TuRBO to manage many decision variables and achieves more than a $2\times$ improvement in the $10\%$-worst user rates over a 3GPP baseline, while also enhancing UAV connectivity by up to $5\times$ in median rates without harming ground users. Transfer learning is demonstrated as a practical mechanism to generalize solutions across related scenarios, enabling convergence with similar iteration counts and minimal performance loss even when limited or no target-data is available. Together, these results highlight the potential of data-driven, transfer-aware optimization for 3D connectivity in urban networks and UAV-enabled applications, with clear pathways for multi-objective extensions and additional decision variables.

Abstract

We propose a data-driven approach for large-scale cellular network optimization, using a production cellular network in London as a case study and employing Sionna ray tracing for site-specific channel propagation modeling. We optimize base station antenna tilts and half-power beamwidths, resulting in more than double the 10\%-worst user rates compared to a 3GPP baseline. In scenarios involving aerial users, we identify configurations that increase their median rates fivefold without compromising ground user performance. We further demonstrate the efficacy of model generalization through transfer learning, leveraging available data from a scenario source to predict the optimal solution for a scenario target within a similar number of iterations, without requiring a new initial dataset, and with a negligible performance loss.

Data-driven Optimization and Transfer Learning for Cellular Network Antenna Configurations

TL;DR

This work presents a data-driven, high-dimensional Bayesian optimization framework to jointly optimize base-station tilts and vertical HPBWs in a real London cellular network, using site-specific 3D ray-tracing with Sionna to model propagation. The approach employs TuRBO to manage many decision variables and achieves more than a improvement in the -worst user rates over a 3GPP baseline, while also enhancing UAV connectivity by up to in median rates without harming ground users. Transfer learning is demonstrated as a practical mechanism to generalize solutions across related scenarios, enabling convergence with similar iteration counts and minimal performance loss even when limited or no target-data is available. Together, these results highlight the potential of data-driven, transfer-aware optimization for 3D connectivity in urban networks and UAV-enabled applications, with clear pathways for multi-objective extensions and additional decision variables.

Abstract

We propose a data-driven approach for large-scale cellular network optimization, using a production cellular network in London as a case study and employing Sionna ray tracing for site-specific channel propagation modeling. We optimize base station antenna tilts and half-power beamwidths, resulting in more than double the 10\%-worst user rates compared to a 3GPP baseline. In scenarios involving aerial users, we identify configurations that increase their median rates fivefold without compromising ground user performance. We further demonstrate the efficacy of model generalization through transfer learning, leveraging available data from a scenario source to predict the optimal solution for a scenario target within a similar number of iterations, without requiring a new initial dataset, and with a negligible performance loss.

Paper Structure

This paper contains 8 sections, 12 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. System Model
  3. Data-driven Optimization
  4. Introduction to Bayesian Optimization
  5. High-dimensional Bayesian Optimization
  6. Case Studies
  7. Transfer Learning
  8. Conclusion

Figures (6)

  • Figure 1: A section of the area considered, with cell deployment sites indicated by black markers and 3D aerial corridors shown in blue.
  • Figure 2: Rates achieved by GUEs and UAVs when the network is optimized for GUEs only (case study #1), GUEs and UAV corridors (case study #2), and with a 3GPP baseline configuration.
  • Figure 3: Optimal tilts and HPBW for case study #1, with GUEs only.
  • Figure 4: Optimal tilts and HPBW for case study #2. Green circles and blue diamonds denote BSs serving GUEs and UAVs, respectively.
  • Figure 5: Convergence of transfer learning applied on case study #2. The initial dataset $\mathcal{D}$ contains $N_{\textrm{o}}=200$ observations.
  • ...and 1 more figures