Aisle-End Session Drops in a High-Bay Distribution Center
Northline Distribution retrofitted a 180,000 ft² grocery DC with 22 ceiling-mounted Wi-Fi 6 access points (4x4) at 38 ft, mounted above the racking rather than in the aisles. The design used 80 MHz channels on 5 GHz, 40 MHz on 2.4 GHz, and every radio at maximum transmit power. Predictive coverage showed -58 dBm everywhere at 30 ft above ground level, and the as-built validation survey, walked at floor level with a passive scanner, passed every KPI. Two weeks into production, 140 handheld picking scanners lose their session at aisle ends and re-associate 8–15 seconds later; 12 autonomous mobile robots (5 GHz only, no 802.11r support) stall in aisles 12–16 and need manual recovery. Operations is raising 40-plus tickets per shift. The integrator proposes raising power another 3 dB and adding six APs on the same 80 MHz plan. The site cannot shut down; changes must be made in 90-minute windows between shifts.
1. Diagnose the most likely root cause of the scanner drops and robot stalls, and explain why the integrator's proposed fix would worsen the problem.
Read the worked answer
Three design decisions interact. First, mounting APs 38 ft up above the racking puts the antenna main lobe at the top of the racks; the energy reaching pick height (4–6 ft) is sidelobe and steep-angle energy, so a passive survey at 30 ft looked excellent while the real client elevation is 10–20 dB weaker and shadowed whenever a pallet or a metal rack face is in the path. Second, 80 MHz on 5 GHz leaves only two non-overlapping channels (UNII-1 and UNII-3), so roughly 11 APs share each channel at full power: oversized cells, sticky clients, high noise floor and retries, and no clean roam target. Third, maximum power with no minimum-RSSI or 802.11k/v support makes the scanner hold its AP until the link fails, then perform a full reauthentication; robots without 802.11r add the 8–15 second stall. Adding APs on the same 80 MHz plan increases co-channel overlap and retries without adding usable capacity, so it makes the symptom worse rather than better. The fix is narrower channels, lower power, aisle-level APs and faster roaming.
2. Define the validation methodology and acceptance criteria you would use before and after the remediation, and explain what the original as-built survey missed.
Read the worked answer
Re-run the survey as an active survey using the actual production client models, at pick height along aisle centrelines and at rack faces, logging RSSI, SNR, retry rate and client-side roam events (association time, whether 802.11r or OKC was used, deauth reason codes). Suggested acceptance criteria: at least -67 dBm and 25 dB SNR on 5 GHz at pick height in every aisle; no more than three APs above -82 dBm on the serving channel; roam completion under 200 ms with under 1% packet loss on a scripted walk; scanner sessions stable across a continuous 60-minute pick run; channel utilization measured at shift start with the full fleet associated. The original survey answered 'is there signal at 30 ft', not 'can 140 clients hold sessions at pick height'. Passive floor-level walks with a laptop adapter also hide the client-side roaming behaviour that is actually failing.
3. Sequence the remediation inside 90-minute windows, and state what you would cut over first, what you would monitor, and what the rollback looks like.
Read the worked answer
Change one variable at a time so you can attribute the result. Window 1: reduce channel width from 80 MHz to 20 MHz on 5 GHz and 20 MHz on 2.4 GHz, and lower transmit power to target -67 dBm at pick height. This needs no cabling or new hardware and is the largest single win, because it multiplies available channels and shrinks oversized cells. Window 2: disable the lowest rates (2.4 GHz 1/2/5.5/11 Mb/s, 5 GHz 6/9/12 Mb/s) and enable 802.11k/v where the scanners support them. Window 3: pilot aisle-mounted APs with directional downtilt in aisles 12–16, the area where robots fail, before committing to the full build. Before each window, export the running configuration per AP group and confirm the controller backup restores; the rollback is a config restore, not a re-survey. Monitor retries, association failures, roam times and channel utilization for the first 15 minutes of the following shift with an on-floor walk test, and revert if retries rise against the baseline.