Simulation dataset for "Joint Electrical and Mechanical Antenna Tilt Optimization for LTE Downlink Networks: Robust-Optimal Policies, Multi-Cell Coordination, and Multi-Terrain Validation at Scale"

Simulation Data and Code - Zenodo Repository

Author: Tanvir AhmedAffiliation: Gdańsk University of Technology, PolandDOI: 10.5281/zenodo.20059153Associated paper: "Joint Electrical and Mechanical Antenna Tilt Optimization for LTE Downlink Networks: Robust-Optimal Policies, Multi-Cell Coordination, and Multi-Terrain Validation at Scale"

Repository Contents

File
Description

lte_tilt_validation.m
Main MATLAB simulation script

Results_20260506_224149.csv
Full simulation grid results (325 tilt configurations × 4 terrains × 500 MC iterations)

Findings_Summary.txt
Per-terrain optimal tilt configurations

run_log.txt
Console output from the simulation run

Ablation_Results.csv
Ablation study: No Tilt, No Averaging, No ML conditions

Baseline_Mechanical_Sweep.csv
Mechanical-only tilt sweep (Urban terrain, elec=0°)

CellEdge_Performance_Summary.csv
5th-percentile throughput and SINR per terrain

EE_Sensitivity_Grid.csv
Energy efficiency over alpha–beta parameter grid

Geo_Validation_CI.csv
Geo-derived validation RMSE and 95% confidence interval

ML_CrossValidation.csv
4-fold cross-validation MAE per fold

ML_Feature_Importance.csv
Linear model coefficient magnitudes per terrain feature

ML_Training_History.csv
Neural network training MSE per epoch

MultiCell_Cluster_Results.csv
Three-scenario cluster throughput (S1 uniform, S2 greedy, S3 coordinated)

Performance_Comparison_Table.csv
Comparison against four prior studies

PL_Model_Curves.csv
Path loss vs distance for Hata, COST231, and WINNER-II models

Power_Model_Sensitivity.csv
Energy efficiency sensitivity to beta parameter

Robustness_Analysis_Results.csv
PF and RR throughput vs shadow fading standard deviation

Statistical_Significance.csv
Per-terrain improvement percentage and Cohen's d effect size

Requirements

MATLAB R2020b or later

Statistics and Machine Learning Toolbox

Neural Network Toolbox (Deep Learning Toolbox)

OpenCelliD tower coordinate files (see Data Sources below)

How to Run

Place lte_tilt_validation.m in a working directory.

Place the OpenCelliD CSV files (poland_towers.csv, denmark_tower1.csv, denmark_tower2.csv) in the same directory. See Data Sources below.

Run the script in MATLAB. All outputs are written to a timestamped subdirectory created automatically.

Runtime is approximately 60–90 minutes on a standard desktop (Intel Core i7, 16 GB RAM).

Simulation Parameters

Parameter
Value

Carrier frequency
2100 MHz

System bandwidth
10 MHz

BS antenna height
30 m

UE height
1.5 m

Cell radius
3.0 km

Transmit power
38 dBm

Monte Carlo iterations
500

UEs per cell
30

MIMO spatial streams
2

Shadow fading std dev
10 dB (terrain-specific in sensitivity sweep)

Mechanical tilt range
0–6° in 0.5° steps

Electrical tilt range
0–12° in 0.5° steps

Total grid size
325 configurations

Terrains
Urban, Suburban, Hilly, Vehicular

Path Loss Models

Three propagation models are averaged as an ensemble:

Hata (1980): Urban macrocell, 150–1500 MHz extended to 2100 MHz

COST231 (1999): Extension of Hata for 1500–2000 MHz

WINNER-II C2 (2008): Urban macrocell scenario, valid 2–6 GHz; PL = 26·log₁₀(d_m) + 20·log₁₀(f_GHz) + 46.8

Path loss model curves across 0.1–10 km are exported to PL_Model_Curves.csv.

Geo-Derived Validation

Tower coordinates from OpenCelliD are used to compute inter-site distances, from which empirical path loss values are derived and compared against the Hata model. This produces a geo-derived validation RMSE of 1.64 dB (95% CI: [1.455, 1.832] dB) across 11,764 coordinate pairs.

Note: The geo-validation uses distance geometry derived from tower locations only, not measured received power. It validates the path loss model's distance-dependent behaviour, not absolute power levels.

Notes on Specific Output Files

Robustness_Analysis_Results.csv

The robustness sweep evaluates the Urban optimal configuration (mech=5.5°, elec=1.5°) across shadow fading standard deviations of 4–16 dB. The Proportional Fair (PF) scheduler result reflects multi-user diversity: with 30 UEs and higher shadow variance, PF gains from occasionally scheduling users in transient strong-signal conditions. The Round-Robin (RR) scheduler result reflects the mean single-user experience without diversity exploitation. The two schedulers are not directly comparable in absolute throughput terms; they represent upper-bound (PF) and lower-bound (RR) scheduling performance respectively.

CellEdge_Performance_Summary.csv

The 5th-percentile throughput metric reflects the worst-case users in the simulation. With a 3.0 km cell radius, 30 UEs uniformly distributed, and 10 dB shadow fading, the lowest-percentile users consistently fall below the CQI-1 SINR threshold (−6.7 dB), resulting in a throughput floor of 0.78 Mbps (CQI-1 × 10 MHz × 2 streams × 0.65 overhead × 0.4 PRB load). This floor is a consequence of the simulation geometry and shadow variance parameters, not a modelling error. Mean throughput and median SINR, which are the metrics reported in the paper, are unaffected.

Performance_Comparison_Table.csv

This table lists four prior studies alongside this work for comparison. Throughput values for prior studies are taken from the respective publications as cited in the paper. NaN entries indicate metrics not reported in those studies.

Data Sources

OpenCelliD (tower coordinates used for geo-derived validation):OpenCelliD Project, "Open database of cell tower locations," https://opencellid.org/ (accessed October 2025).Files required: Poland LTE towers, Denmark LTE towers (two regional extracts).These files are not included in this repository due to their size. They can be downloaded directly from the OpenCelliD website under the Creative Commons Attribution-ShareAlike 4.0 licence.

Citation

If you use this code or data in your research, please cite the associated paper:

T. Ahmed, "Joint Electrical and Mechanical Antenna Tilt Optimization for LTE Downlink Networks: Robust-Optimal Policies, Multi-Cell Coordination, and Multi-Terrain Validation at Scale," 2026.

And the data repository:

T. Ahmed, "LTE Joint Tilt Optimization Simulation Data," Zenodo, 2026, doi: 10.5281/zenodo.20059153.