France Data Curation¶

This document explains how the French scenario is assembled. France is the only country in the set with two independent base-station fields: a crowdsourced one and an official one, which lets the simulator quantify how much a result depends on the source rather than on the network.

Raw Data Sources¶

• data/france/opencellid/208.csv: Cell locations for MCC 208.
  • Source: OpenCellID, crowdsourced. 390,737 records, of which the four national operators hold: Orange 116,994 (208-01), Bouygues 94,931 (208-20), Free Mobile 87,410 (208-15), SFR 86,758 (208-10).
• data/france/anfr/208.csv: Official site locations, derived from the ANFR BNIR monthly export.
  • Source: ANFR, Donnees sur les installations radioelectriques de plus de 5 watts. Every operator is legally required to declare each installation above 5 W, so this is the French counterpart of the FCC Antenna Structure Registry used for the United States.
• Communes with population and boundaries: fetched at build time from geo.api.gouv.fr, the Etalab service that republishes INSEE legal population and IGN commune contours. No manual download is needed.

Processing Scripts¶

data/processing_scripts/standardize_france.py¶

Builds the population layer. It

1. Fetches all communes with their legal population and centroid.
2. Excludes overseas departments (codes starting 97 and 98), keeping metropolitan France including Corsica, so that the simulated field is one contiguous landmass. This mirrors the mainland cut applied to Portugal.
3. Fetches commune contours department by department and simplifies them to a 200 m tolerance, which is invisible to the rejection sampling used for agent placement but keeps the polygon file at 19.5 MB instead of roughly 250 MB.
4. Writes ids with leading zeros stripped, because the loader normalizes any integer-looking id through parse(Int, .) and would otherwise fail to join "01001" against 1001.
5. Outputs municipalities.csv (34,746 communes, 66,165,815 inhabitants), regions.geojson and cities.csv.

data/processing_scripts/standardize_france_anfr.py¶

Builds the official base-station field. It joins SUP_SUPPORT (support geometry, coordinates in degrees, minutes and seconds) to SUP_STATION (station to operator), maps the ANFR operator id to the MNC used by OpenCellID, converts coordinates to decimal degrees, drops anything outside metropolitan France, and writes the result in the 14-column OpenCellID layout the loader expects.

Unzip the monthly ANFR export into data/france/agent-unprocessed-raw-datasets/anfr/ before running it.

Cells Are Not Sites¶

The two fields describe the same network at different granularity, and the difference is large enough to matter:

An OpenCellID record is a cell, so a three-sector site appears roughly three times, and because each position is independently estimated from crowdsourced measurements, those sectors land hundreds of metres apart rather than on top of each other. An ANFR record is a site. The simulator treats every row as one gNB, so the OpenCellID field produces a denser apparent deployment and a higher handover rate than the physical network has.

Neither field is wrong; they answer different questions. Run the OpenCellID field for comparability with the other countries, and the ANFR field when the absolute handover rate has to correspond to real sites.

Scenario Parameters¶

julia --project main.jl national france