UniFi · Wi-Fi performance · Fix
Wi-Fi is slow and clients cling to the wrong AP — stop changing settings, start measuring
Step 0 — measure first (three instruments, no settings)
- Wired baseline: speed-test from a cable at the switch (WiFiman's device→gateway test does the same job). Slow wired = the WAN/gateway is your disease; this page no longer applies.
- The walk: WiFiman's Signal view, room by room: RSSI, band, serving AP — and stand on the seam between two APs and watch when (if) you roam. Ubiquiti's own scale: above −60 excellent · −60…−70 acceptable · −70…−75 “clients should begin to roam” · below −80 instability.
- The AP-side look: Radios → Coverage shows per-client signal and the AP/Client Signal Balance — a “Poor” balance is the shouting-AP/whispering-phone asymmetry made visible. AirView shows two bars: orange = interference, blue = your own airtime. Read historically with Ubiquiti's key: high airtime + high interference = channel disease; high airtime + low interference = load/multicast disease.
Name your disease
| Evidence | Disease | Ladder rungs |
|---|---|---|
| Fine near APs, dead in specific rooms regardless of settings | Coverage hole | Placement/AP — not settings (§FAQ) |
| AirView orange bar high; neighbours on your channel | Co-channel interference | 1–2 |
| Clients at −75 hanging on, near AP visible but <8 dB better | Sticky client (power asymmetry) | 3, then 4 |
| Airtime high, interference low; many chatty devices | Airtime hog (multicast/broadcast + legacy rates) | 5 |
| Fast near the gateway AP, slow behind the mesh node | Uplink ceiling | Wire it (Pack #23's ≥50%/hop law) |
| Wired baseline already slow | WAN | Gateway/ISP guides, not this one |
The settings ladder — in order
- Channels & widths. 2.4 GHz: 20 MHz, channels 1/6/11 only. 5 GHz: 40 MHz in offices, 80 at home (Ubiquiti's own guidance — its speeds page pushes 80 while its connectivity guide says 40 for offices; the goal decides, and neighbouring networks vote too). 6 GHz can go wide (160–320) where spectrum is clean. DFS caveat from Ubiquiti's text: on a radar hit the AP stops broadcasting on that channel immediately, boot/switch waits run 1–10 minutes, and a detected channel is off-limits ≥30 minutes — a DFS channel that “keeps vanishing” is behaving legally. Verify: AirView orange bar falls. Revert: note the old plan first.
- One plan, all APs: adjacent APs on different channels; same channel on neighbours is the interference you just measured.
- Transmit-power symmetry. High in low-density, Medium in offices (Ubiquiti) — and above all matched to client reality: your phone transmits far weaker than your AP. Lowering the shouting far AP is what creates the 8–12 dB margin the client's roam logic demands. Verify: the seam test roams earlier; Signal Balance improves.
- Steering & roaming assists. Band Steering stays enabled (Ubiquiti; now SSID-level via BSS transition). BSS Transition on: it “encourages” roam-ready clients (old devices may ignore it). Fast Roaming (802.11r): Ubiquiti “strongly recommends” on — with its two printed caveats: it conflicts with Switch Port Isolation, and clients that can't do 802.11r belong on a separate legacy SSID with it off. Verify: walk-test calls/Teams through the seam.
- Airtime controls. Multicast & Broadcast Control and Minimum Data Rate Control are the documented levers (raising minimum rates trims multicast/broadcast airtime and “encourages clients to roam sooner”; Ubiquiti warns 802.11b-era devices may lose connectivity — they were the airtime problem). Multicast Enhancement is the gentler first step. Note: older guides prescribe an “Airtime Fairness” toggle; it does not appear in Ubiquiti's current settings documentation — this ladder teaches what exists.
- Minimum RSSI — last, surveyed, gentle. Only with survey numbers in hand, per-AP/per-radio, starting at −80 dBm (Ubiquiti's starting point, unchanged across versions) — and never at folklore values: −67 dBm boots an iPhone before its own −70 trigger starts scanning. If rung 3 fixed the margin, this rung stays empty.
What NOT to do
| The reflex | Why it backfires |
|---|---|
| Max transmit power everywhere | Manufactures sticky clients and co-channel noise; the phones can't shout back |
| 320 MHz “because Wi-Fi 7” | One giant channel in shared spectrum = you become everyone's interference |
| Minimum RSSI −67/−70 from a forum | Sits at/above client roam triggers — drops, not roams |
| Ten settings at once | Unattributable results, loops included (see the provisioning guide) |
| Chasing the negotiated-rate number | Ubiquiti: real throughput ≈ 50–60% of PHY rate; 150–600 Mbps per client on gigabit is realistic |
FAQ
All settings perfect, one room still dead.
Then it was never a settings problem: coverage is geometry. Move the AP, add an AP (wired if humanly possible — mesh rules), or accept the room's physics. No toggle bends radio waves around a masonry wall.
Android/Windows clients — same thresholds?
Their vendors don't publish roam thresholds; anyone quoting a universal “−70 law” beyond Apple/Ubiquiti's corroborated range is embroidering. The margin principle (near AP must win decisively) holds across clients even where exact numbers don't.
One ancient device slows everyone — real?
Yes, qualitatively: airtime is shared, and slow-rate transmissions (legacy clients, distant clients) occupy more of it; that's the documented mechanism behind rung 5. No exact “one device = X% slower” multiplier exists in Ubiquiti's material, so this guide doesn't print one.
Sources
Ubiquiti Help Center — WiFi Troubleshooting Guide, Airtime Utilization, Maximizing Wireless Speeds, DFS Channels, SSID & AP Settings Overview (signal scale, widths, power, DFS behaviour, steering/roaming toggles, airtime controls, Minimum RSSI stance) · Apple — Wi-Fi deployment reference (roam triggers −70/−75 dBm, 8/12 dB candidate margins) · This site's Packs #21/#23/#25 (6 GHz conditions, mesh law, Channel AI clearance).
Did this fix work for you?
Your reaction tells us whether this guide solved the problem it promised to solve.