A Wi-Fi survey belongs before the ceiling goes up

A predictive Wi-Fi survey is a model built from a floor plan: walls, materials, ceiling height, and a set of assumptions about what is behind them. It is genuinely useful, it costs almost nothing, and it is the right first step. It is also a model, and the building will differ from it.

What the drawing does not show

The drawing shows a wall. It does not show that the wall is a double layer of plasterboard on a steel frame with a foil-backed insulation board in the middle, which is a great deal more attenuation than plasterboard.

It does not show the racking. Warehouse plans are drawn empty, and a warehouse is not empty — it is a grid of steel uprights holding metal shelves loaded with stock whose density changes weekly. A survey done in an empty building is a survey of a building that will never exist again.

It does not show the mirrors in the bathroom corridor, the lift shaft, the fire doors that are held open by magnets and close during a drill, or the plant room somebody decided to put a repeater in.

Two surveys, not one

The pattern that works is predictive first, then a validation survey once the space is built and before the ceiling is closed.

The second one is the one that saves money, because the cost of moving an access point is entirely determined by whether the ceiling is open. Before the tiles go in, moving a point three metres is twenty minutes. After, it is a cable pull through a live ceiling void, usually out of hours, and somebody has to make good.

Density beats coverage

The most common design mistake is optimising for coverage — a signal everywhere — when the actual requirement is capacity, meaning enough radios that the clients in one area are not all contending for the same one.

A single high-powered access point covering an open-plan floor produces a perfect coverage heat map and a terrible experience at 3pm, because every one of those clients is sharing one radio and the slowest of them sets the pace. Several lower-powered points, each covering less, is almost always the better answer.

This is also why turning the transmit power up is rarely the fix it appears to be. It extends the cell of the access point, so more clients join it, and it does nothing for the client's own transmit power — the laptop still cannot shout back from the far corner. You get a device that shows four bars and cannot pass traffic.

Survey with the right device

Survey with the worst client you have to support, not the best. A current laptop with a good radio will report a usable signal in places a handheld barcode scanner will not, and if the scanners are what the business runs on then the scanner is what the design has to satisfy.