Warehouse Wi-Fi and the Scanner Dropouts

The symptom points to a roaming failure at the cell edge, made worse by co-channel interference, not to an AP or controller outage.

Evidence:

• Duration. A 3–8 second stall is the signature of a client that has lost the link and is scanning every channel before reassociating - plus possible DHCP or application re-authentication on top. A true AP reboot or PoE fault would produce a longer, synchronized outage across many scanners.
• Location pattern. Drops cluster at cross-aisles. Cross-aisles are where a client physically crosses a cell boundary, so it is exactly where insufficient overlap causes a hard roam.
• Signal levels. -79 dBm with 18 dB SNR at aisle ends is below the point where a 2.4 GHz-only client stops trying to hold on. Most such clients wait until roughly -80 dBm before scanning, which is too late for a clean handoff.
• Mounting. APs above 24 ft of steel racking radiate over the top of the racks. Signal reaching the aisle floor is a strong reflection/leakage path with poor spatial consistency, so coverage looks fine at the AP and is thin down in the aisle.
• Channel reuse. Twelve APs across three channels at 20 MHz is four APs per channel with heavy overlap - high co-channel interference drives retries and slows the roam scan.

Placement and coverage

• Move APs off the rack tops and into the aisles, mounted on rack uprights or drop-poles at roughly 15–20 ft, so the Fresnel zone runs down the aisle instead of over the steel.
• Design for -67 dBm at the cell edge on the target band and 15–20% overlap at aisle intersections - enough for the client to see the next AP before it loses the current one.
• Use directional or downtilt antennas in long aisles to stretch coverage along the aisle rather than wasting it into racking.

2.4 GHz plan for the scanners

• Keep 20 MHz and channels 1/6/11 only, but re-run the plan with the new placement so reuse distance increases.
• Disable the low rates (1, 2, 5.5, 11 Mbps). Removing them shrinks the effective cell and pushes clients to roam earlier rather than clinging to a distant AP at 1 Mbps.
• Enable a minimum RSSI / band-steering-adjacent setting so the controller disassociates clients below about -75 dBm, if the scanner vendor supports fast reconnection.
• Turn off aggressive load balancing if enabled - it deauthenticates clients mid-roam and would add seconds to the drop.

5 GHz plan for laptops

• Fixed, non-overlapping channel plan at 40 MHz; leave the 2.4 GHz band to the scanners only rather than band-steering laptops onto it.

Client side

• Confirm the scanners' roaming aggressiveness setting; many handhelds have a configurable threshold. If they are end-of-life, budget replacement with dual-band units - that is a design fix, not a workaround.

Measure, don't declare victory from a survey map:

• Roam time per event from the controller's client history - target sub-500 ms from last frame on AP A to first frame on AP B. Anything still in the seconds is a client-side or DHCP problem, not coverage.
• Retry and error rates per AP and per band, before and after, taken at the same time of shift. A drop in retries confirms the interference and cell-size changes actually took effect.
• SNR distribution sampled by walking the aisles, not by standing under an AP.
• DHCP and authentication latency for a scanner that re-associates - if the 3–8 seconds is mostly a DHCP discover timeout, the problem is the wired path, not RF.
• Application-level session logs from the warehouse system, since that is the business metric.

Reopen the design if dropouts persist only in specific aisles (placement still wrong), only at shift change when hundreds of clients are present (capacity, not coverage), or only after a controller failover (mobility tunnel or config drift). Also reopen it if fixing coverage shifted the problem to 5 GHz laptops - a design change on one band frequently disturbs the other.