Six floors, 78 access points, and a channel plan that was doing the damage

The firm had a well-documented problem and an expensive theory about it. Video calls degraded from around 09:30, recovered at lunchtime and degraded again at 14:00. Users described the network as "fine when nobody is in".

Over eighteen months, IT had responded by adding access points. The count had gone from 40 to 78. Each round of additions produced a brief improvement followed by a return to the same complaints, and the most recent round had made things measurably worse on two floors.

By the time we were engaged, the working assumption inside the business was that the building was somehow hostile to Wi-Fi and that a different vendor was required.

What the data said

Two days of controller statistics settled it. Signal strength was excellent almost everywhere — median client RSSI around -58 dBm. Coverage was not the problem and had not been for some time.

Peak 5 GHz channel utilisation was running at 78 percent during business hours. Above 70 percent, users reliably describe a network as broken regardless of how strong the signal is.

Three configuration decisions were producing it:

• 80 MHz channels throughout. With DFS disabled, that left four usable 5 GHz channels for 78 access points.
• DFS switched off — a decision taken years earlier, by someone no longer at the firm, for reasons nobody could recall.
• Every radio at maximum transmit power, including 2.4 GHz on all 78 access points across three channels.

Each addition of access points had increased the number of radios sharing four channels. The firm had been paying to make the problem worse.

The work divided into a decision, a reconfiguration and a reduction. Only the last one required a ladder.

Resolving the DFS question

We could not justify re-enabling DFS from a map alone, so we measured. A single access point was parked on each candidate DFS channel and left for fourteen days with radar events logged from the controller.

The result: four events across the fortnight, all on channels 120–128, none on 52–64 or 100–116. The building is far enough from Heathrow's terminal approach and well away from coastal weather radar.

Decision: enable DFS on channels 52–64 and 100–116. Permanently exclude 120, 124 and 128. This took the channel budget from 4 usable 80 MHz channels to roughly 12 at 40 MHz — a threefold increase before a single other change.

Reconfiguration

• 5 GHz channel width reduced from 80 MHz to 40 MHz. Peak per-client throughput fell; aggregate capacity rose substantially. In a firm whose heaviest real workload is video conferencing and document management, that is the correct trade.
• Transmit power reduced to 14 dBm at 5 GHz and 8 dBm at 2.4 GHz, matched to the weakest client class in the estate — a fleet of older IP handsets.
• Minimum basic rate raised to 24 Mbps, disabling 802.11b rates entirely. This shrank effective cell sizes and stopped clients clinging to distant access points.
• 2.4 GHz radios disabled on 44 of the remaining access points, leaving 18 covering the building for IoT, handsets and a small number of legacy devices.
• Automatic channel selection constrained to the approved channel list and fixed at 40 MHz, rather than being left free to choose any width.
• Multicast rate-limited and disabled entirely on the guest network.

Reduction

With the channel plan corrected, 16 access points were surplus — they were adding contention without adding coverage. Rather than leave them powered, we removed them and kept them as spares, which also recovered PoE budget on two switches that had been close to their limit.

Removing access points required more explanation to the business than any other part of the project.

Validation

A full post-deployment survey on the live configuration, walked floor by floor, at 5 GHz and 6 GHz separately. Roaming tested with a real handset on a real call, walking the route from reception to the sixth floor, twice.

Peak channel utilisation fell from 78 percent to 31 percent. Mean client SNR improved by 10 dB — not because signal got stronger, but because the noise floor fell once fewer radios were shouting over each other at maximum power.

Measured aggregate throughput per access point more than tripled, despite the channel width being halved. This surprised the firm's IT team more than any other number, and it is the clearest illustration of what co-channel contention actually costs.

Video call drop reports fell from 34 per week to 2. The two remaining were traced to a specific laptop model with a known driver issue and resolved separately.

What did not go to plan

Enabling 802.11r fast roaming broke a fleet of older IP handsets on day one. We had tested it — but on the current-generation handsets, not the 2019 units still in use in two departments. The handsets were moved to a separate SSID without fast transition while they are replaced.

That mistake cost half a day and is the reason our current practice is to test roaming features against the oldest device in an estate first, not the newest.