Wi-Fi 8 Multi-AP Coordination, Explained for People Who Deploy APs

I've spent twenty-some years hanging access points in convention centers, stadiums, school districts, and — during one very long sixty days in 2020 — across an entire city. And in all that time, the fundamental assumption of Wi-Fi design has never changed: your neighbor's AP is your enemy. Every co-channel cell you can hear is contention you have to eat. We channel-plan around it, we power-tune around it, we survey around it. Wi-Fi 8 is the first amendment that looks at that assumption and says: what if the AP next door was on your side?

That's Multi-AP Coordination — MAPC — the headline feature of IEEE 802.11bn, the amendment behind Wi-Fi 8. The working group calls the whole effort UHR, "Ultra High Reliability," and that name tells you everything about the goal: not faster peak rates, but predictable behavior in dense, ugly, real-world RF. As of this writing (July 2026), the D2.0 draft is going to working-group letter ballot at the Montreal plenary, silicon is already announced, and the marketing machine is spinning up. So let's separate what's actually in the draft from what's hype, because there's a lot of both.

What is Multi-AP Coordination in Wi-Fi 8?

MAPC is a framework that lets neighboring APs — including APs in different BSSs, on the same channel — cooperate on how they use airtime instead of blindly deferring to each other through CSMA/CA. The mechanics matter here, because the first misconception I hear is "so it's centrally scheduled like LTE now." No. CSMA/CA is still underneath everything. What changes is what happens after an AP wins a TXOP.

Under 802.11bn, the AP that wins EDCA contention becomes the coordinating AP (the sharing AP), and it can donate part of the TXOP it just won to neighboring coordinated APs. It's opportunistic sharing of airtime the winner already owns — not a central scheduler handing out grants. Before any of that happens, APs go through MAPC Discovery, negotiate MAPC Agreements (which can be updated or torn down), and optionally authenticate the inter-AP relationship using PASN so the coordination channel itself is secured. That last bit matters if you've ever thought about what a rogue AP could do with a protocol that lets one AP tell another to lower its transmit power.

Per the current drafts (D1.x, and everything I'm describing is keyed to the excellent Wilhelmi et al. tutorial on Draft 1.4 — the single best MAPC reference in existence right now), five coordination modes made the cut: Coordinated Spatial Reuse (Co-SR), Coordinated TDMA (Co-TDMA), Coordinated Beamforming (Co-BF), Coordinated Restricted TWT (Co-RTWT), and Coordinated Channel Recommendation (Co-CR). Let's walk through them like people who actually have to make this stuff work.

How does Coordinated Spatial Reuse work?

Co-SR is the mode I'd bet on mattering first, and it's easiest to understand as the redemption arc of a feature most of us turned off. If you deployed 802.11ax, you remember OBSS-PD spatial reuse: raise your preamble-detect threshold, pretend you can't hear your neighbor, transmit anyway, and hope. It was a blind gamble on a static threshold, and in most enterprise deployments I touched, it stayed disabled because "hope" is not a link budget.

Co-SR replaces the gamble with a negotiation. Downlink only. The coordinating AP that won the TXOP sends a Co-SR Invite — carrying the minimum and maximum OFDM symbols, whether an ICF/ICR exchange is needed, and the exchange duration. The coordinated AP answers with a Co-SR Response, there's an optional ICF-ICR exchange, and then the coordinating AP fires a Co-SR Trigger that confirms the agreed parameters — including transmit power. And here's the part that changes everything: the coordinating AP explicitly sets both its own Tx power and the maximum allowed Tx power of the coordinated AP. Both APs then transmit downlink in parallel, with mutual interference that's been managed on purpose, per TXOP, instead of assumed away by a threshold somebody set in a GUI two years ago.

Two flavors exist: Mode 1 admits both UHR and EHT (Wi-Fi 7) PPDUs; Mode 2 is UHR-only. What the draft deliberately does not specify is the algorithm for computing those power levels. That's vendor secret sauce, and it's where the real differentiation — and the real interop headaches — will live.

Does it work? The simulation data is genuinely impressive, with genuinely important caveats. A July 2025 study out of UPF ran 4-AP enterprise scenarios and saw 99th-percentile delay reductions of 31–95% versus plain DCF — the best case cut 127.5 ms down to 6.24 ms at 20-meter inter-AP spacing, with aggregate throughput up to +284% in favorable geometry. But read the fine print like an engineer: gains shrink at 10-meter spacing, shrink with bursty traffic, and sparse networks see almost nothing. And it's all simulation — no public hardware measurements exist yet. Keep that in your pocket for when the datasheets arrive.

What are the other four MAPC modes?

• Co-TDMA — the workhorse, no downlink/uplink restriction. Three phases: the coordinating AP polls neighbors with an ICF inside a BSRP Trigger, then hands out slices of its TXOP via MU-RTS TXS Trigger frames (AP ID, allocated time, channels — the coordinated AP answers CTS and transmits), and any unspent time comes back via a MAPC TXOP Return frame. One elegant rule: donated time can only carry traffic of equal or higher priority than the access category that won the TXOP, so you can't win a voice TXOP and hand it to your neighbor's bulk downloads. Think of it as micro time-slicing one TXOP across neighboring BSSs.
• Co-BF — the flashy one, downlink only. Cross-BSS sounding lets each AP steer a null toward the other AP's client while beamforming to its own, so two APs transmit simultaneously with spatial separation. It's also the most PHY-demanding mode, and the draft caps it hard (per D1.x — hedge accordingly until D2.0 settles): 2 APs max, 2 spatial streams per STA, 4 total spatial streams, 4 recipient STAs across both APs. This is null-steering between a pair of APs, full stop.
• Co-RTWT — cross-BSS protection for latency-sensitive service periods. Your AP advertises Restricted TWT schedules in its beacons; a coordinated neighbor translates them into its own TSF timebase and ends its transmissions (or defers backoff) before your protected service period starts. If you run AR/VR, robotics, or voice on R-TWT, this stops the AP next door from stomping your protected windows.
• Co-CR — APs exchange profiles and advertise recommended channels for peer-to-peer traffic, pushing things like screen mirroring off your contested service channels. Anyone who's watched a Miracast session eat a classroom channel understands why this exists.

Equally important is what's not in 802.11bn. Joint Transmission — multiple APs coherently transmitting to one client, the "distributed MIMO / cell-free Wi-Fi" future everybody's slideware promised — is not in the draft. Neither is Coordinated OFDMA. Both got punted to future amendments, the same way multi-AP itself got punted from 11be. If someone tells you Wi-Fi 8 does cell-free MIMO, they're describing a research paper, not the standard.

Wi-Fi 7 vs Wi-Fi 8: what actually changes?

The honest answer: nothing about the ceiling, everything about the floor. Same three bands (2.4/5/6 GHz), same 320 MHz max channel, same ~23 Gbps theoretical max as Wi-Fi 7. The UHR targets are +25% throughput at the same SNR, −25% latency at the 95th percentile, and −25% MPDU loss — including during roaming. That last one comes via the Seamless Mobility Domain (SMD) framework: a set of AP MLDs under an SMD Management Entity where a client keeps its association and auth state across APs — context transferred ahead of the move, new link activated, old AP releases state, no reassociation, no full 4-way re-handshake. Make-before-break, built on MLO. There's also a PHY reliability kit — ELR long-range modes, distributed-tone RUs (dRU) that beat 6 GHz LPI PSD caps for up to ~11 dB of effective power gain, unequal modulation per stream, four new MCSs — and NPCA/DSO, which let devices use non-primary sub-channels when the primary is jammed with OBSS traffic. Samsung Research's 802.11bn deep-dive and MediaTek's Wi-Fi 8 whitepaper cover these well.

Capability	Wi-Fi 6/6E (802.11ax)	Wi-Fi 7 (802.11be)	Wi-Fi 8 (802.11bn)
Design goal	Efficiency (OFDMA)	Extreme throughput	Ultra High Reliability
Max channel width	160 MHz	320 MHz	320 MHz (unchanged)
Theoretical peak	~9.6 Gbps	~23 Gbps	~23 Gbps (unchanged)
Bands	2.4/5 (+6 for 6E)	2.4/5/6 GHz	2.4/5/6 GHz (unchanged)
Multi-AP behavior	OBSS-PD (blind threshold)	None (punted from 11be)	MAPC: Co-SR, Co-TDMA, Co-BF, Co-RTWT, Co-CR
Inter-AP power control	No — static PD threshold	No	Yes — per-TXOP negotiated Tx power (Co-SR)
Roaming	802.11r/OKC	802.11r/OKC (+MLO within one AP MLD)	SMD: cross-AP context transfer, make-before-break
Busy-primary relief	No	Preamble puncturing	NPCA + DSO (use non-primary sub-channels)
Cell-edge tools	—	—	ELR, dRU, unequal modulation, 4 new MCSs
Status (mid-2026)	Ratified, certified	Ratified, certified	D2.0 ballot Jul 2026; ratification May 2028; WFA cert Jan 2028

What does MAPC mean for AP density and site surveys?

Here's where I want to talk to the people who actually carry ladders. MAPC's entire premise is making dense, overlapping cells cooperative instead of adversarial — and that quietly inverts a design instinct we've spent two decades sharpening.

Co-channel neighbors become coordination candidates. MAPC coordination happens between APs that can hear each other on the same channel. A channel reuse plan that perfectly isolates every co-channel cell — the thing we've all been graded on — will partially defeat MAPC, because APs that can't hear each other can't coordinate. The design question shifts from "minimize cell overlap" to "engineer useful overlap."

It is not a license to thin out APs. I need to say this twice because a VAR will eventually say the opposite in a bid. Coordination mitigates the cost of overlap; it does not create coverage. SNR still sets the floor, Co-BF is capped at two APs and four total streams, and the simulation data shows sparse networks gain almost nothing while the sweet spot is moderate overlap — roughly that 20-meter-spacing regime — with gains shrinking again when you cram APs at 10 meters. Density done right gets rewarded. Density done wrong still hurts. Sparse still starves.

Surveys gain a new measurement. Still SNR-first, always. But inter-AP path loss — which APs actually hear which, and how well — now defines Co-SR and Co-TDMA group feasibility, which makes it a design input, not trivia. And fair warning: as of WLPC Phoenix 2026, no survey tool ships Wi-Fi 8 modeling. The tooling will lag the silicon, same as it did with 6 GHz.

The unspecified parts are the vendor parts. Coordinating-AP selection, Co-SR power computation, MAPC group formation — all outside the standard. The frames are standard; the brains are proprietary. Expect single-vendor MAPC first, no cross-vendor interop demos as of today, and the Wi-Fi Alliance certification in January 2028 to define the baseline interoperable subset. SMD is likewise an infrastructure-side architecture — where the SMD-ME lives is a controller/cloud design decision, and it supersedes the 11r/OKC patchwork only within one vendor's mobility domain. Plan for bigger L2 mobility domains again, and keep 11r for the legacy fleet.

Legacy clients get roughly nothing directly. Co-SR Mode 2 is UHR-only, Mode 1 admits EHT, SMD needs UHR MLD clients. Your existing fleet benefits indirectly from reduced contention, and that's it. Client refresh cycles gate the payoff, same as every generation.

What would I do today? (Deployment guidance, July 2026)

When we built out McAllen's citywide network — a thousand-plus APs and 24 CBRS base stations in sixty days — the thing that hurt us was never peak throughput. It was contention behavior under real density: co-channel cells fighting each other at park boundaries, roaming transitions dropping packets exactly when a kid's remote-learning session could least afford it. Wi-Fi 8 is the first standard aimed squarely at that pain. But aiming isn't shipping, so here's my honest guidance for mid-2026:

• Don't wait to buy what you need now. If your Wi-Fi 5/6 gear is due, buy Wi-Fi 7 without guilt. Enterprise Wi-Fi 8 APs realistically land mid/late 2027 into 2028, per Qualcomm's Rolf de Vegt's timeline, and certified interop arrives January 2028. Consumer gear ships earlier — late 2026 into 2027 — pre-certification, with firmware risk to match.
• Read datasheets, not box logos. Broadcom's CES lineup includes dual-band Wi-Fi 8 chips with no 6 GHz radio. The Wi-Fi 7 bifurcation lives on: "Wi-Fi 8" on the box guarantees nothing about the feature set.
• Design today's networks MAPC-ready. Deploy at proper density for coverage and capacity — don't over-cram, don't thin out. When MAPC firmware arrives, moderate co-channel overlap is an asset, so stop treating every co-channel neighbor at −78 dBm as a design failure. Document inter-AP path loss on your next survey; it's tomorrow's coordination-group map.
• Get your fleet onto MLO now. SMD roaming is built on MLO machinery and needs UHR clients. Every Wi-Fi 7 MLD client you deploy in 2026–2027 is a client that's one hop from the good stuff. If roaming loss is killing you today, tighten your 11r/OKC config — SMD won't rescue you before 2028.
• Hold vendors to specifics. Ask which MAPC modes they'll ship at launch, whether their numbers come from hardware or simulation (today the honest answer is simulation only), what their Co-SR power algorithm assumes, and where the SMD-ME lives in their architecture. Anyone promising cross-vendor MAPC in 2026 is ahead of reality — no one has demonstrated it.
• Latency-sensitive verticals: pay attention early. Warehousing, healthcare, AV-heavy venues — Co-RTWT plus SMD plus P-EDCA is aimed at you, and the 95th/99th-percentile latency numbers are where UHR earns its name. Track the official TGbn timeline and start piloting when D2.0-frozen enterprise silicon appears in 2027.

The big picture: Wi-Fi 8 won't win a speed test against Wi-Fi 7, and it isn't trying to. It's trying to make the worst five percent of your network's moments — the roam that drops packets, the cell edge that crawls, the co-channel neighbor that stomps your voice traffic — boring and predictable. After twenty years of designing around adversarial APs, I'm genuinely happy to start designing around cooperative ones. Details will keep shifting through D3.0 and D4.0 comment resolution into 2027, but the headline features are set. The APs are finally learning to play nice. About time.