Exeter Coverage Checker

How it works

• Cells and masts. Each drive-test reading lists every operator's four strongest cells (PCI and EARFCN) with their RSRP. For each cell, a path-loss curve RSRP = A − 10·n·log10(d) is fitted to its readings, binned to 20 m squares, which estimates where its mast is. Positions from road readings are approximate, typically a few hundred metres off.
• Hills and buildings. From each mast, rays sample the Environment Agency LiDAR terrain every 20 m; a point behind a hill loses signal by knife-edge diffraction, more at 3.5 GHz than at 800 MHz. Buildings and trees enter as clutter: the mean height of everything within about 150 m, with a loss per metre fitted to the drive tests for each operator.
• Coverage. On a 25 m grid, each operator's coverage is its strongest modelled cell, shifted by the operator's median difference from the measurements. Near the roads the remaining measured − modelled difference is spread out over about 150 m, so the map matches the drive tests there. Away from them a smooth random field (shadow fading, ±4 dB for 4G and ±6 dB for 5G, mostly shared between operators) stands in for what the LiDAR cannot explain, so coverage thins out unevenly with distance, as it does in reality. It is one fixed draw, not a measurement.
• 5G. Most 5G readings are not tied to a cell, so 5G is modelled the way it is rolled out: a 5G cell on each of the operator's estimated masts, at its 5G frequency (EE mostly 700 MHz), calibrated to the measured 5G. VMO2 and Vodafone 5G was not scanned in Exeter, so theirs is synthetic: 3.5 GHz on all their estimated masts, calibrated to the 3.6 GHz 5G measured on Three's network (now part of VodafoneThree).
• Residents. Census 2021 residents were spread to every address (UPRN) in an earlier project; each address takes the modelled coverage of its 25 m square. This page only holds residents per 25 m square, not per address, so a drawn area counts the squares whose centre falls inside it.
• Not operator coverage. Mast positions are estimates, there are no antenna patterns, and the drive tests are from 2022–2024. Use it as an estimate from open data, not as an operator's coverage map.
• Validation. One in five 1 km squares is held out, the whole model is rebuilt without its readings, and those squares are predicted. Errors are in dB against the measured strongest-cell RSRP.