The warehouse where the survey was done with empty racking

A distribution centre commissioned eighteen months earlier, signed off against a clean post-installation survey, was losing scan transactions at shift change. Operators described handhelds "hanging" mid-pick and having to walk to the end of an aisle to complete a transaction.

The operator had the original survey. The heat maps were green. They were also, as it turned out, entirely accurate for the building as it had been measured — empty.

What had changed

Nothing about the network. Everything about the building.

The original survey was conducted during fit-out, before racking was installed and before any stock arrived. Eleven metre racking, loaded with mixed palletised goods — a substantial proportion of it metal stock and liquid in drums — attenuates 5 GHz dramatically. Where the empty building had been a large open box with excellent propagation down every sight line, the loaded building was a grid of absorbent canyons.

The ceiling-mounted omnidirectional grid, appropriate for the empty box, was illuminating the tops of racks and the roof space. At floor level in the aisles, where every handheld actually operates, coverage was patchy.

The second factor

Handheld scanners are unforgiving clients. Small antennas, low transmit power, aggressive power-save behaviour and — in this estate — a device model with conservative roaming thresholds that would cling to an access point down to around -85 dBm rather than roam.

Marginal coverage plus sticky clients produces exactly the symptom reported: a device that appears connected, sitting at the lowest data rate, taking most of a second to complete a roam and timing out the transaction in the process.

Re-survey, loaded

The first and most important step was a survey of the building as it actually operates: full racking, normal stock mix, during a working shift. We surveyed at aisle floor level with a representative handheld, not with a survey adapter — the adapter has a materially better antenna than the device that has to work.

That survey produced a coverage map bearing little resemblance to the commissioning document, with eleven distinct areas below -70 dBm at handheld height.

Re-architecting the RF

The design changed from a ceiling grid to an aisle-aligned directional layout:

• Directional patch antennas mounted at approximately 7 m, aimed down the aisles rather than down from the roof.
• Access points positioned to illuminate along aisles, using the racking as the waveguide it has become rather than fighting through it.
• Seven additional access points, concentrated in the cross-aisles and the two deep-pick zones that had the worst coverage.
• 2.4 GHz retained across the estate at low power, because a proportion of the handheld fleet is dual-band but prefers 2.4 GHz in this environment — and here, unusually, that is the right outcome.

Client-side work

• Minimum basic rate raised to 12 Mbps to shrink cells and encourage roaming.
• 802.11k neighbour reports enabled, which the handheld firmware does support and was not being given.
• Roaming thresholds adjusted on the handhelds themselves via the device management platform — the single highest-impact change, and one the original project never touched because it is on the client, not the network.
• A firmware update to the handheld fleet, which included a roaming fix in the release notes that nobody had read.

Validation

Walked the full pick route at floor level with a scanner running a continuous transaction loop, at shift change, with the building working normally.

Scan transaction timeouts fell from around 61 per shift to 3. Coverage at -65 dBm or better along aisle floors went from 62 percent to 97 percent. Mean roam time dropped from 890 ms to 120 ms — the client-side threshold change accounting for most of that on its own.

The operator's own measure of success was simpler: pickers stopped walking to the end of aisles to complete transactions.

The part worth noting

Seven additional access points were needed. Thirty-four had been installed originally. Had the original survey been conducted with racking loaded, the correct design would have been apparent at commissioning and the antenna choice would have been directional from the start.

The remediation cost more than the difference would have, because it happened in an operational building at night rather than during fit-out.