Wi-Fi RF propagation

Predictive Wi-Fi has a reputation problem, largely earned by tools that draw circles around dots and call it coverage. This page describes what is actually computed.

The hybrid path-loss selection

Two models run for every point, and the more conservative answer wins.

The first is discrete: the path from access point to point is traced through the building, and every wall it crosses is charged its own attenuation. The second is statistical: the ITU-R P.1238 indoor model, which characterises an environment with a path-loss exponent rather than individual obstructions.

Taking the greater of the two is a deliberate choice. Discrete alone flatters open-plan space, because a path that happens to cross no wall looks like free space when in practice scattering and clutter are eating signal. Statistical alone flatters dense space, because an averaged exponent cannot know about the concrete core you are standing behind. Choosing the more pessimistic of the two means neither a sparse office nor a dense warehouse comes out optimistic.

Materials

Per-material, per-band attenuation values are held for each construction type across 2.4, 5 and 6 GHz, sourced against NIST IR 6055 measurements. Drywall, glass, concrete and masonry each carry their own figure per band, which matters because attenuation is not flat across spectrum — the 6 GHz band pays more for the same wall than 2.4 GHz does, and a design validated at one band is not automatically valid at another.

Reflection is handled with ITU-R P.2040 complex permittivity and Fresnel coefficients rather than treating every surface as a perfect absorber.

Diffraction

Signal bends around obstructions, and a model that ignores it will under-predict coverage behind structure. Three mechanisms are implemented: knife-edge diffraction, the Uniform Theory of Diffraction, and Deygout multiple-edge diffraction per ITU-R P.526 §4.5.4 for paths crossing several obstructions.

Fresnel-zone clearance is evaluated alongside, because a path that is geometrically clear but has an obstruction inside the first Fresnel zone is not actually clear.

Warehouse racking

Racking is charged as a function of how much of it the signal crosses, not as a binary wall. Attenuation scales with rack depth and with the angle of incidence — a path cutting diagonally through a rack run traverses more steel than one crossing perpendicular. Both the depth multiplier and the angle multiplier are bounded, so a pathological geometry cannot produce an unbounded loss figure.

This is the single largest difference between a warehouse prediction here and one from a tool that treats racking as furniture.

Multi-floor

Floor penetration is applied per ITU-R P.1238 by construction type and band, so a design that leans on coverage bleeding down from the floor above is evaluated honestly rather than assumed.

Calibration

The model can be corrected against reality. Feed in measured signal readings from the building and the engine correlates predictions against them and fits per-wall correction factors, which then apply across the prediction engines.

This is the honest middle ground between pure prediction and a full measured survey: you are not required to walk the building, but if you have walked part of it, that evidence improves the model rather than sitting in a separate report.

One calculation chain

The heatmap, the placement preview, the auto-planner, the walkthrough readout and the remediation suggestions all run the same calculation chain. That is a correctness property rather than an implementation detail — it means the number you see when you hover a point is the number the planner optimised against, not a second approximation drawn for display.