Coordinated Spatial Reuse in Wi‑Fi

• Coordinated Spatial Reuse (Co-SR) is a Wi‑Fi 8 mechanism that allows multiple access points or AP–STA pairs to transmit simultaneously by explicitly coordinating interference levels.
• It leverages techniques ranging from controller-assisted scheduling and TXOP sharing to online learning over packet detection thresholds, transmit power, and MCS selection.
• Studies report throughput gains up to 280% over legacy methods, highlighting its potential to improve performance in dense WLAN deployments while managing delay and fairness.

Coordinated Spatial Reuse (Co-SR, also written C-SR or CSR) denotes a family of IEEE 802.11bn / Wi‑Fi 8 Multi-Access Point Coordination (MAPC) mechanisms in which multiple APs or AP–STA pairs are allowed to transmit simultaneously on the same channel under explicit coordination, provided that interference remains acceptable for the participating links (Wilhelmi et al., 2023). In the recent Wi‑Fi literature, Co-SR is positioned as a response to dense overlapping basic service set deployments, where legacy distributed contention and purely local spatial reuse rules leave substantial concurrency unrealized. Its realizations range from controller-assisted group formation and TXOP sharing to online learning over Packet Detect (PD), transmit power, station selection, and MCS selection (Wilhelmi et al., 2024).

1. Conceptual definition and relation to legacy spatial reuse

Within the 802.11bn MAPC agenda, Co-SR is the coordinated mechanism that increases the likelihood of simultaneous transmissions while suppressing incompatible ones (Wilhelmi et al., 2023). The defining distinction is that reuse is not left to independent carrier-sensing decisions. Instead, APs are treated as coupled decision makers whose transmission opportunities, power levels, or reuse groups are selected with knowledge of interference relations, scheduling constraints, or learned feedback (Wojnar et al., 12 May 2025).

This separates Co-SR from two older baselines that recur throughout the literature. In legacy Distributed Coordination Function (DCF), APs contend independently through CSMA/CA, which in dense topologies leads to excessive deferral and access serialization (Wilhelmi et al., 2023). In 802.11ax Overlapping Basic Service Set Packet Detect (OBSS/PD), a device may ignore weaker energy from another BSS if the sensed power is below a threshold, but that remains a local threshold-based mechanism rather than a cooperative one (Wilhelmi et al., 2023). Several Wi‑Fi 8 studies therefore treat Co-SR not as an incremental CCA relaxation, but as a change in access paradigm from random independent access or local threshold tuning to coordinated concurrent access (Nunez et al., 2024).

A second recurring distinction concerns what exactly is being coordinated. In the PD/TPC formulation of AI-native spatial reuse, the action of each AP can be a joint SR operating point,

$$a \in \mathcal{A} = \{(\mathrm{PD}_1, P_1), (\mathrm{PD}_2, P_2), \dots\},$$

where $\mathrm{PD}_i$ is a packet-detect threshold setting and $P_i$ is a transmit-power level (Wilhelmi et al., 2024). In TXOP-sharing formulations, by contrast, the coordinated object is a set of AP–STA pairs admitted into the same coordinated slot, often with accompanying power and MCS assignments (Nunez et al., 2021). This suggests that Co-SR is best understood as an umbrella for coordinated reuse policies rather than as a single fixed MAC procedure.

2. Coordination architectures and protocol realizations

The most common architectural setting is MAPC, in which APs exchange coordination information, jointly adapt actions, and use performance feedback to steer later transmissions (Wilhelmi et al., 2024). Several papers assume a central controller connected over a wired backbone that collects cross-AP information, determines compatibility for simultaneous transmission, and schedules coordinated slots periodically on top of ordinary Wi‑Fi operation (Nunez et al., 2023). In that framework, coordinated MAPC transmissions occur every $T$ ms, while the intervals between them remain uncoordinated CSMA/CA “breathing” periods (Nunez et al., 2023).

An important line of work implements Co-SR through TXOP sharing. When a coordinated AP wins contention, it becomes the sharing AP, reserves the channel, and triggers one or more compatible AP–STA pairs from other BSSs to transmit during the same coordinated TXOP (Nunez et al., 24 Jul 2025). Over-the-air procedures are described with MAP-RTS, simultaneous MAP-CTS replies from shared APs, and MAP-TF frames that allocate coordinated slots and signal parameters such as MCS, bandwidth, and TXOP duration (Nunez et al., 2023). In 802.11be-era precursor work, this appeared as c-TDMA/SR: c-TDMA supplies the slot structure, while coordinated spatial reuse allows several APs to occupy the same coordinated slot when interference permits (Nunez et al., 2021).

The operational granularity differs across studies. Some formulations are group based: a Co-SR group is a set of APs or AP–STA pairs that may transmit together if they are mutually compatible under a SINR or capture constraint (Nunez et al., 2023). Others are controller-driven and hierarchical: after discovery or pairing, one AP in a coordinated group acts as the sharing AP, the others as shared APs, and the sharing AP may be selected in round-robin fashion (Chen et al., 21 Mar 2026). Still others preserve fully distributed contention and add a coordination phase only after one AP has already acquired a TXOP, yielding what one study characterizes as a compete-then-collaborate design (Yu et al., 17 Jun 2025).

Simulation platforms reflect this architectural diversity. Komondor is used for coordinated MA-MAB evaluation of SR parameter control (Wilhelmi et al., 2024), while Kom8ndor extends the Komondor line with Wi‑Fi 8 MAPC features including Co-SR, Co-TDMA, and Co-BF (Wilhelmi et al., 24 Jun 2026). The available Kom8ndor description identifies Co-SR as a first-class MAPC feature, but does not disclose a full Co-SR state machine or detailed coordination algorithm in the excerpted material (Wilhelmi et al., 24 Jun 2026).

3. Group formation, interference constraints, and control variables

At the core of Co-SR is the compatibility problem: determining which simultaneous transmissions remain decodable under mutual interference. In analytical and protocol papers alike, compatibility is expressed through a receiver-side SINR condition. One Wi‑Fi 8 scheduling framework requires that for a candidate set of APs, all associated stations must observe sufficient SINR when those APs transmit simultaneously, and the resulting group is admitted only if the feasibility condition holds for all involved links (Nunez et al., 2023). Another latency-focused study states the same idea through a capture-threshold criterion, under which AP–STA pairs can transmit simultaneously only if the receiver SINR remains above $\gamma_{\textrm{CE}}$ (Nunez et al., 24 Jul 2025).

Once feasibility is defined, the literature diverges on how to construct groups. A practical heuristic is the At-most-$K$ method, which treats each AP as a reference group head and greedily adds up to $K-1$ other APs, preferring candidates with the lowest RSSI impact at the reference AP’s stations and checking mutual SINR compatibility after each addition (Nunez et al., 2023). In another line of work, candidate combinations of AP–STA pairs are scored by how many packets they are expected to deliver and then assembled into a group set under a fairness-style appearance constraint so that each pair belongs to a limited number of scheduled groups (Nunez et al., 2024). A plausible implication is that Co-SR group creation is structurally close to clique or packing problems, even when papers adopt heuristics rather than explicit graph-theoretic optimization.

Control variables also differ by formulation. In coordinated SR parameter tuning, the knobs are PD threshold and transmit power (Wilhelmi et al., 2024). In scheduling-centric models, the variables are the concurrent AP set, the destination station served by each participating AP, and optionally the transmit power used in that TXOP (Wojnar et al., 7 Jan 2025). In more holistic MAPC control, transmit powers are discretized, link adaptation selects an MCS from the 802.11 20 MHz MCS set, and the binary scheduling variable $a_{i,j,k}$ indicates whether AP $j$ serves STA $i$ in TXOP $\mathrm{PD}_i$0 (Chen et al., 21 Mar 2026). This broader treatment makes Co-SR a joint optimization of concurrency, power, and PHY operating point rather than only a reuse admission problem.

The same issue appears in machine-learning schedulers validated against upper bounds. There, the C-SR scheduling problem is not merely “which APs transmit concurrently,” but also “which station each transmitting AP serves,” “what transmit power each AP should use,” and “which MCS can be supported under the resulting SINR” (Wojnar et al., 12 May 2025). That formulation is especially important because several studies report that larger coordinated groups are useful only if SINR remains high enough to avoid a collapse in the selected MCS (Nunez et al., 2023).

4. Formal models and algorithmic formulations

The formal treatment of Co-SR spans analytical performance models, exact or upper-bound optimizations, online bandit learning, and decentralized game-theoretic learning. An analytical strand models the feasible concurrent transmission states directly. In the CTMC approach, the system state is the set of APs transmitting at a given time, transitions correspond to transmission start and end events with rates $\mathrm{PD}_i$1 and $\mathrm{PD}_i$2, and throughput follows from the stationary occupancy of feasible states,

$\mathrm{PD}_i$3

with spatial efficiency measured as a normalized form such as $\mathrm{PD}_i$4 (Wilhelmi et al., 2023). A related analytical line extends Bianchi’s DCF model so that a successful slot can carry multiple simultaneous transmissions rather than only one, thereby embedding group-selection probabilities and MAPC overhead into aggregate throughput expressions (Nunez et al., 2024).

Optimization-oriented work constructs upper bounds or controller decisions explicitly. A mixed-integer linear programming framework uses column generation to build feasible transmission sets and then allocate normalized time shares $\mathrm{PD}_i$5 to optimize either aggregate throughput or the throughput of the worst-served station (Wojnar et al., 12 May 2025). Centralized hierarchical MAB formulations expand the control space further, choosing scheduling, association, power, MCS, and QoS target $\mathrm{PD}_i$6 through a two-layer decision process whose objective may be a weighted sum of total data rate and Jain’s fairness index, or proportional fairness under minimum-rate constraints (Chen et al., 21 Mar 2026).

A major contemporary strand formulates Co-SR as a multi-agent multi-armed bandit problem. In coordinated SR parameter control, APs are agents, SR configurations are arms, and rewards are local or shared network-performance signals. The key complication is that each agent’s reward depends on the actions of the others, so the environment is non-stationary from every local perspective (Wilhelmi et al., 2024). Hierarchical MABs address the combinatorial size of the action space by splitting group selection into levels: first selecting which APs join the sharing AP, then selecting a recipient station for each selected AP (Wojnar et al., 7 Jan 2025). A deeper hierarchy adds transmit-power choice as a further level, producing a three-level structure that selects concurrent APs, stations, and powers (Wojnar et al., 12 May 2025).

More elaborate reinforcement-learning approaches convert Co-SR into a hierarchical multi-agent control problem. One fully distributed HMARL design decomposes the process into a polling phase and a decision phase; a high-level policy chooses the station to serve during the TXOP, a low-level policy chooses transmit power for each packet transmission, and compressed inter-AP messages are exchanged through a neural encoder (Yu et al., 17 Jun 2025). Another centralized framework uses a two-layer MAB to update QoS targets over a slower outer horizon and power, association, and MCS choices in an inner loop (Chen et al., 21 Mar 2026).

The literature also contains an important decentralized counterpoint. Earlier work on dense, uncoordinated wireless networks showed that channel and transmission-power choices can be learned selfishly through $\mathrm{PD}_i$7-greedy, EXP3, UCB, and Thompson sampling, with sequential learning reducing temporal throughput variability relative to concurrent action updates (Wilhelmi et al., 2017). More recently, internal regret minimization has been proposed as a decentralized alternative to MAPC-heavy coordination: instead of converging to inefficient Nash equilibria under external-regret learning, competing BSSs are guided toward correlated equilibria, thereby mimicking coordination without explicit communication (Wilhelmi et al., 9 Feb 2026). This debate connects Co-SR to an older theoretical lineage in spatial congestion games and Aloha games with spatial reuse, where distributed users coordinate implicitly through potential-game dynamics or fixed-point adaptation under topology-specific interference graphs [(Chen et al., 2012); (Lyu et al., 2013)].

5. Reported performance across analytical, simulation, and testbed studies

The quantitative literature on Co-SR is large but methodologically heterogeneous. Reported gains come from CTMC analyses, Bianchi-style models, Monte Carlo simulators, Komondor-based evaluations, MATLAB studies, Python/PyTorch simulators, and an openwifi SDR testbed (Wilhelmi et al., 2023, Wilhelmi et al., 2024, Wojnar et al., 12 May 2025). The figures therefore describe scenario-specific outcomes rather than a single universal operating point.

Study	Setting	Reported outcome
(Wilhelmi et al., 2023)	CTMC throughput analysis	up to 59% over legacy DCF; up to 42% over 802.11ax OBSS/PD
(Nunez et al., 2024)	Bianchi-based 4-AP analysis	throughput gains from 54% to 280%
(Wilhelmi et al., 2024)	Coordinated MA-MAB in Komondor	mean throughput +15%; minimum throughput +210%; maximum access delay below 3 ms
(Nunez et al., 24 Jul 2025)	4-AP MATLAB study	delay reduction from 31% to 95% versus DCF
(Wojnar et al., 12 May 2025)	ML scheduling, simulation and testbed	H-MAB improves aggregate throughput over legacy IEEE 802.11 by 80% on average in random scenarios
(Chen et al., 21 Mar 2026)	6-AP two-layer MAB	Jain fairness index 0.709 baseline, 0.92 with proportional fairness, 0.97 with weighted sum
(Nunez et al., 2021)	c-TDMA/SR in random deployments	throughput gains higher than 140% in 90% of four-AP scenarios

Several cross-paper patterns are explicit. Gains are largest when topology admits larger compatible concurrent groups, AP separation is sufficient to keep mutual interference manageable, or coordinated power control enables simultaneous transmissions that uncoordinated methods would reject (Wilhelmi et al., 2023, Nunez et al., 2024). Delay benefits are especially pronounced in loaded networks, where serving multiple queues per contention win reduces waiting time before service (Nunez et al., 24 Jul 2025). Fairness effects depend strongly on reward design and scheduler structure: per-AP backlog-aware schedulers outperform per-group schedulers for 95th-percentile delay (Nunez et al., 2023), and fairness-aware reward terms or objective functions materially change the resulting AP balance (Chen et al., 21 Mar 2026, Yu et al., 17 Jun 2025).

The implementation evidence is narrower than the simulation evidence but nontrivial. In the Ghent University Industrial IoT Lab testbed using openwifi SDR-based devices, both flat MAB and H-MAB quickly found configurations enabling three concurrent transmissions, and more than 80% of chosen actions involved three parallel transmissions (Wojnar et al., 12 May 2025). This does not remove the gap between idealized models and deployment-scale Wi‑Fi 8 systems, but it shows that Co-SR is not only an analytical construct.

6. Limitations, misconceptions, and open research directions

A common misconception is that Co-SR is merely a coordinated variant of OBSS/PD. The literature instead treats it as a broader coordinated reuse family that may involve group creation, TXOP sharing, power adjustment, MCS selection, buffer-aware scheduling, or online learning (Nunez et al., 2023, Wilhelmi et al., 2024). Another misconception is that Co-SR is a single protocol. In practice, papers analyze periodic controller-assisted transmissions, DCF-based over-the-air coordination, hierarchical MAB schedulers, PPO-based distributed cooperation, and fully decentralized internal-regret methods (Nunez et al., 2024, Yu et al., 17 Jun 2025, Wilhelmi et al., 9 Feb 2026).

The principal technical tension is between coordination quality and coordination cost. Centralized or MAPC-heavy approaches can optimize across multiple coupled parameters, but several papers note signaling overhead, negotiation overhead, architectural complexity, and dependence on backhaul or controller support (Chen et al., 21 Mar 2026, Nunez et al., 24 Jul 2025). Decentralized approaches reduce overhead and sometimes question the need for heavy signaling, but must cope with severe non-stationarity, imperfect local information, and weaker guarantees on global operating points (Wilhelmi et al., 9 Feb 2026, Wilhelmi et al., 2017).

Standardization direction is another constraint. Some Wi‑Fi 8 studies explicitly distinguish an unconstrained grouping mode from a MAX2 mode because current TGbn discussions are described as initially supporting at most two simultaneous transmissions, even though larger groups can yield substantially better delay and throughput in favorable topologies (Nunez et al., 24 Jul 2025). This suggests a gap between near-term standard conservatism and the performance potential reported in research models.

Methodological limitations are also explicit. Many results are simulation based; one coordinated MA-MAB paper states that over-the-air validation and standard-compatible implementation remain important next steps (Wilhelmi et al., 2024). Group-creation heuristics leave parameter selection open, as with the empirical choice of $\mathrm{PD}_i$8 in At-most-$\mathrm{PD}_i$9 grouping (Nunez et al., 2023). Learning-based methods can require retraining or lose performance under major topology mismatch (Yu et al., 17 Jun 2025). Even simulator support is uneven: Kom8ndor identifies Co-SR as a key MAPC feature for Wi‑Fi 8 research, but the available description does not expose a full detailed Co-SR state machine or coordination message sequence (Wilhelmi et al., 24 Jun 2026).

Taken together, the literature presents Co-SR as a central Wi‑Fi 8 mechanism for exploiting dense-network concurrency, but not as a solved problem. The stable core is clear: simultaneous multi-AP transmission under explicit interference-aware coordination. The unsettled questions concern how that coordination should be implemented, how much signaling it should require, how broadly it should optimize across MAC and PHY knobs, and whether explicit MAPC control or implicit decentralized adaptation will prove the more scalable route in dense WLANs (Wilhelmi et al., 2023, Wilhelmi et al., 9 Feb 2026).