Commissioning Wireless in a Building Full of Metal

Coverage predictions are made from a floor plan. Radio waves do not see a floor plan. They see steel racking eight metres high, a stainless tank that reflects almost perfectly, a mesh mezzanine deck that behaves like a partial ground plane, and a forklift that drives through the middle of the link every ninety seconds.

This is why wireless in industrial buildings behaves differently from wireless in offices, and why survey practice that works in an office fails on a plant floor. An office is mostly plasterboard, glass and people. A warehouse is mostly metal and air, and metal does two things simultaneously: it blocks, and it reflects. The blocking is intuitive. The reflecting is what causes the problems nobody expects.

Multipath is the mechanism. A signal arrives at the receiver by several routes of different lengths — direct, plus one bounce off the racking, plus another off the roof deck. Those copies arrive slightly out of step and can partially cancel. The result is that signal strength varies enormously over distances of a few centimetres, so a device mounted where the installer found a convenient bracket may sit in a null while an identical device half a metre away works perfectly. It also means a strong signal reading is not the same as a good link; modern radios report signal-to-noise and error rates, and those are the numbers that predict whether a connection will hold.

The environment changes, which surveys rarely capture. Racking that is full of product in October is a different radio environment from racking that is empty in January. A liquid-filled tank absorbs strongly; the same tank empty does not. Doors open and close. Vehicles move. A survey conducted on a quiet Sunday when the plant is shut down measures a building that does not exist during production, and that single choice invalidates more surveys than any technical error.

So survey when it is busy, or at least twice. Walk the actual paths devices will use, at the actual heights they will be mounted, with the actual antennas — a survey done with a laptop at chest height tells you very little about a sensor bolted to a machine frame at ankle height. Record signal-to-noise rather than signal strength alone. And spend time on the edges of coverage rather than the middle, because the middle is fine everywhere and the edges are where the fleet will live.

Interference deserves separate attention on 2.4 GHz, which remains crowded in ways that surprise people. Industrial microwave heating, older cordless equipment, Bluetooth devices, and the plant's own Wi-Fi all share the band, and a spectrum analyser sweep — even a cheap one — will show sources that no amount of channel planning fixes. Where the application allows it, 5 GHz or a sub-GHz technology sidesteps the problem entirely. Where it does not, knowing what else is transmitting is the difference between engineering and guessing.

Two habits worth adopting. Mount for radio first and convenience second, or at least know which one you traded; a bracket moved thirty centimetres to clear a null costs nothing at installation and a great deal afterwards. And write down what was measured where, with dates. When the network starts misbehaving in eighteen months, the question will be what changed, and a baseline is the only way to answer it.

One more thing, learned the hard way by most people who do this: test with the machine running. A press, a drive, a welder or a large contactor can raise the noise floor enough to break a link that was solid during a quiet survey, and that failure will be intermittent, correlated with production, and blamed on the software.