Distribution Centre: Evaluating an 867 Mbps-per-AP Design Claim

867 Mbps is a PHY rate, not throughput. It corresponds to 802.11ac, 80 MHz, 2 spatial streams, MCS9 (256-QAM, 5/6 coding), which requires roughly 30 dB SNR and a clean channel. Neither condition holds at cell edge or between steel racks.

Airtime accounting:

• MAC overhead (preamble, IFS, backoff, BA/ACK, beacons, probes) plus retries typically leaves 50–60 % of PHY as TCP goodput: about 430–520 Mbps for one client at close range, and less once 80 MHz is reused.
• 24 APs on 80 MHz cannot each get a clean channel, so medium sharing cuts that figure again; assume 250–350 Mbps per AP as a planning number.
• 40 clients sharing 300 Mbps fairly gives ~7.5 Mbps average, which superficially covers 2 Mbps each - but scanners are bursty, voice needs <50 ms latency and <1 % loss, and fair sharing does not deliver tail latency.
• The configured 1 Mbps basic rate is the hidden killer: beacons, probes and management frames occupy airtime for hundreds of microseconds each, and every retry at low rate is disproportionately expensive.

Verdict: the claim confuses PHY rate with capacity and ignores airtime. For a mix of bursty scanners and voice, design to 20–25 active clients per radio, remove 1/2/5.5/11 Mbps from the basic rate set, and plan capacity on 5 GHz for the headsets rather than loading 2.4 GHz further.

In the 2.4 GHz band there are only three non-overlapping 20 MHz channels (1, 6, 11). Twenty-four radios therefore means about eight APs per channel, and no channel plan can avoid co-channel reuse - only manage it.

Consequences and adaptation:

• Co-channel interference (CCI) is the normal state. Throughput per AP falls roughly in proportion to the number of co-channel APs whose coverage overlaps, so coverage overlap must be engineered down, not up.
• Reduce transmit power and shrink cells so that co-channel APs are physically separated; on a 12 m ceiling over 3 m aisles this argues for mounting lower (racking tops, 5–7 m) and aiming down aisles rather than a uniform high-ceiling grid.
• Keep 20 MHz width on 2.4 GHz. 40 MHz would consume most of the band and is not viable with 24 radios.
• Do not rely on 2.4 GHz for capacity. Put the 40 voice headsets on 5 GHz with band steering, and treat 2.4 GHz as a coverage layer for the legacy scanners only.
• Use scheduled automatic channel selection and power calibration, then verify with an on-site survey rather than trusting the auto plan.

If the scanner population is concentrated in the picking zone, consider fewer 2.4 GHz radios (for example 12) with better placement instead of 24, since additional 2.4 GHz radios add interference without adding usable capacity.

Likely causes, in order of probability:

1. Multipath and shadowing from steel racking. At 12 m mounting height the path to a handheld at 1 m is oblique and passes through rack structures; the Fresnel zone is obstructed. Signal arrives via reflections with deep nulls, so RSSI may look acceptable while SNR and retry rates are poor.
2. Low-rate airtime congestion. With 1 Mbps basic rate, management and retry traffic from 60 scanners consumes airtime, raising latency until the scanner's session timer expires.
3. Roaming failure. Scanners are 2.4 GHz-only; if coverage overlap between APs is thin at the aisle ends, the client holds a dying link instead of roaming.

Changes:

• Re-site APs lower (racking tops, 5–7 m) or use directional antennas aimed along aisles so the first Fresnel zone is clear.
• Set basic rates to 12/24 Mbps and disable 1/2/5.5/11 Mbps as basic; keep them supported only if some scanners genuinely require 802.11b.
• Ensure 15–20 % coverage overlap at the aisle ends and enable smart roaming with an SNR-based trigger.
• Enable band steering and move headsets to 5 GHz.

Acceptance criteria: at 1 m working height in every aisle, RSSI ≥ −65 dBm and SNR ≥ 25 dB; frame retry rate below 10 %; scanner round-trip latency below 200 ms at peak; zero disassociations in a 2-hour peak soak test.