Strong Signal, 40% Retries in the Distribution Centre

Strong RSSI with heavy retries means the frames are being damaged or deferred by co-channel interference, not by weak signal. SNR is not the limiting factor; CIR is.

Channel arithmetic: 38 APs across 3 non-overlapping channels = ~12–13 APs per channel in one open building. At 20 dBm through ceiling omnis, each cell easily reaches several aisles in every direction, so a receiver routinely hears several APs and dozens of clients at comparable levels. Under CSMA/CA that produces:

• Deferral - the radio waits for energy it does not need to hear, burning airtime it never gets back.
• Collisions - simultaneous transmissions from stations that cannot hear each other.
• Hidden nodes - with a big AP cell and small handheld transmitters, two handhelds can be invisible to each other while both are audible to the AP.

Every collision is a retry; every retry is airtime. At 40% retries the effective throughput is roughly halved, which is exactly what voice cannot tolerate.

Also note the survey was done in an empty building. Full racking, stock, forklifts and 200 people change the RF environment substantially - that plot is not the RF environment you are troubleshooting.

More AP power increases the size of every cell, which increases co-channel overlap on a floor that already has 12–13 APs per channel. You also raise the hidden-node probability: the AP's coverage grows faster than the handheld's 14–15 dBm transmit range, so the gap between 'can hear the AP' and 'can be heard by the AP' widens.

It also worsens client stickiness. The uplink is unchanged, so the handheld at -78 dBm still thinks the distant AP is fine, and that association now looks even more attractive.

What to change instead:

1. Turn power down, e.g. 11–14 dBm, and add APs to restore coverage. Small cells, low power, high reuse.
2. Directional antennas down the aisles rather than omnis above them - racking is a natural waveguide, and aisle-shaped coverage does not bleed into the next aisle.
3. Disable 1/2/5.5/11 Mbps rates to shrink the effective cell and stop slow frames consuming airtime.
4. Enable 802.11k/v so handhelds get neighbour reports and can be steered; check whether 802.11r is supported before enabling it.
5. Set a roaming threshold on the handhelds themselves - the client decides when to roam, and the client is the sticky party.
6. Move voice to 5 GHz if the next handheld refresh allows it; 2.4 GHz with three channels is structurally wrong for dense voice.

Do not re-run the predictive model or re-read the empty-building survey. Measure on the floor, at the client's height, during a live shift:

• Spectrum analysis in aisle 14 at ~1.2 m, not at the ceiling. Look at channel utilisation (duty cycle) first, then amplitude - and look for non-Wi-Fi interferers that a WLAN survey would never show: wireless cameras, cordless phones, microwave or RF welders, improperly shielded equipment, and any 2.4 GHz telemetry.
• Per-AP and per-client retry, CRC-error and frame-error counters, plus client counts per AP, to see whether the problem is local to one AP/channel or systemic.
• Duplicate/co-channel BSSID counts audible from that aisle, to confirm whether a third or fourth AP is competing.
• Uplink check: what level the AP hears the handheld at, not what the handheld hears the AP at. Sticky-client problems live in the uplink.

Timing matters: forklifts, metal racking full of stock, and 200 moving people change multipath and absorption hour by hour. A measurement at 06:00 in an empty aisle is not a measurement of the problem.