Guide

will help you reproduce the experiments from https://arxiv.org/abs/2507.19653 with proper OSM building heights.

Prerequisites

1. Install Dependencies

pip install -r requirements.txt

2. Copy Dataset

Copy the Rome dataset CSV file to the data directory:

cp /path/to/rome_crop_knn_ready.csv .../data/

Running the Complete Workflow

Step 1: Generate Scene with Proper Building Heights

python generate_scene.py

This will:

• Fetch building data from OpenStreetMap for coordinates (41.861824 to 41.8778842, 12.4525696 to 12.471929)
• Extract proper building heights using multiple methods:
  • Direct height attributes from OSM
  • Building levels with appropriate floor heights
  • Building type-specific defaults with realistic variation
• Generate PLY mesh files for all buildings
• Create the Sionna scene XML file at scenes/rome_scene_with_heights.xml

Step 2: Run Baseline Experiment

python run_experiments.py --experiment baseline

Step 3: Run Optimized Per-BS Experiment

python run_experiments.py --experiment optimized

Step 4: Evaluate KNN Localization Performance

python evaluate_knn.py

Step 5 : Run All Parameter Sweeps

python run_experiments.py --experiment sweeps

This runs comprehensive parameter sweeps covering:

• TX heights: [11, 12, 15, 20, 40, 55]m
• Frequencies: [1.01, 1.05, 1.1, 1.2, 1.3, 1.5, 2, 3.6, 5]GHz
• TX/RX patterns: ["iso", "dipole", "hw_dipole", "tr38901"]
• Ray tracing depths: [1, 2, 3, 4, 5, 6, 7]
• TX/RX azimuth orientations: 0° to 330° in 30° steps

File Outputs

All results are saved in the results/ directory:

• *.npy: Simulation result tensors (shape: [n_bs, n_ue, 4] for RSSI, NSINR, NRSRP, NRSRQ)
• *_config.npy: Experiment configuration files
• knn_evaluation_summary.csv: KNN localization results
• rssi_analysis_summary.csv: RSSI statistics analysis

The framework now properly reproduces the paper experiments with realistic building heights extracted from OpenStreetMap data.