R&D sneak peek: Multi-AP Coordinated Spatial Reuse for Wi-Fi 8

In this post, I would like to describe an exciting take on Multi-AP Coordination (MAPC) leveraging Spatial Reuse (SR). AP coordination has groups of APs communicate with each other to minimise collisions while transmitting at the same time and on the same frequency channel. I recently came across a scientific article by David Nunez et al. in [1], which I believe is worth understanding as it proposes an algorithm to identify coordinated AP groups and simultaneously schedule transmissions of their buffered frames.

What is Spatial Reuse?

Spatial Reuse is not a new concept in wireless communications. If a transmitter is “distant” enough from other transmitters on the same channel, it can consider them background noise when communicating with its intended clients. Think about yourself speaking in English to an audience in a large garden. Other people can do the same 20 yards away from yours and each other’s groups if everyone agrees to be well-behaved. Communication is possible because the sound is affected by Path Loss; the same can be applied to 802.11 wireless communication when considering a model such as the TGax model for Enterprise Scenarios [2] which considers the distance between the transmitter and receiver, the frequency used and the number of walls. The mathematical details are not interesting for understanding the concept, but you may appreciate how being in a harsh RF environment causing strong attenuation can reduce the reuse distance. The picture below shows a client station called STA, being able to hear AP₄ and AP₅ on the same channel used by AP₂ to which it is connected.

AP4 and AP5 are interferers and contribute to the effect of background noise in the communications from AP₂, potentially preventing STA from receiving readable signals. The effect of background noise and interference over the useful signal is described by the following formula for the Signal to Interference plus Noise Ratio (SINR), as seen on the client STA for the transmission from AP₂:

where P₂ and P_j are the received power from AP₂ and the other j APs as seen from STA respectively, and N is the noise power. We cannot alter the N value as it is due to external factors in the environment beyond our control (e.g. the APs of a neighbouring office), so we should keep the sum of the P_j as low as possible. The obvious course of action is not to have APs with higher interference power transmit at the same time as AP₂. We face a new problem: how do we select the group of APs that can transmit at the same time as AP₂?

Detecting Coordinated Groups for SR

The paper defines a “Sharing AP” as the coordinator for a group of “Shared APs.” They also assume that all the APs are within the coverage area of the Sharing AP, which can access the medium and send a Multi-AP Request-To-Send (MAP-RST) to reserve the channel. In case of no collisions on their side, the Shared APs reply with a MAP Clear-To-Send (MAP-CTS), guaranteeing that the clients in the service area will stay silent for the rest of the Transmit Opportunity Sharing (TXOP-sharing) transmission. Once the Sharing AP can access the channel, the shared TXOP is divided into one or more coordinated temporal slots. It also signals which APs should transmit during each slot with a MAP Trigger Frame (MAP-TF). Effectively the authors leverage Time Division Multiple Access (TDMA) as they have the Sharing AP reserving the channel, and they enhance it with SR by selecting only transmitters that interfere the least.

To create groups of APs that can transmit together, the authors define a Central Controller (CC) that knows all the received powers P_j as seen from the STA and can compute the SIND for each AP. They assume that the RF channel is always symmetric so that the power the client receives in transmissions from the AP (downlink) is the same as the AP receives from the client on the way back (uplink). In light of this, all the APs in the same coordinated group of size M must guarantee that their SINR for each of their connected client STAs is above an m threshold so that:

The APs within the same group of K units are identified using a “greedy” strategy called the “At-most-K” algorithm. An AP is selected as a reference, and on each iteration a new AP is added if the formula described above holds and the SINR is still high enough for all the client STAs connected to APs already in the group.

Traffic Scheduling Algorithms

The operation can be repeated using different reference APs to find multiple groups of “compatible” APs. APs in better positions (or trivially far enough) may belong to various groups. Once the groups are identified, the CC must select what groups should transmit. The paper proposes four strategies based on the number of packets buffered in the APs of the group:

1. NumPkSingle: the CC selects the groups containing the AP with the highest number of buffered packets, and then picks the single group with the highest number of buffered packets across all the APs in the group.
2. NumPkGroup: the CC selects the group with the most packets.
3. OldOkSingle: the CC selects the groups containing the AP with the oldest buffered packet, and then picks the single group with the highest aggregate group delay as a sum of the waiting times of the oldest buffered packets across all the APs in the group.
4. OldPkGroup: the CC selects the group with the highest aggregate group delay.

The algorithms prevent starvation between groups when using NumPkGroup and OldPkGroup by dividing the aggregated value by the number of APs in the groups; this is particularly important whenever there are groups with just a few APs and others with a large number of units. The simulation results presented in the paper show that NumPkSingle and OldOkSingle (per-AP selection) outperform the other strategies (per-Group selection) when the TXOP-sharing transmissions are scheduled every 5 milliseconds, and the SINR threshold is 20 dB, high enough to guarantee the use of a high Modulation and Coding Scheme (MCS) in the simulation using 9 APs transmitting on the same channel.

Wrapping up

MAPC is still a hot topic for research, and commercially-available APs will probably use some still-unknown solution. I hope this blog post picked the readers’ interest in what is coming in the future iterations of IEEE 802.11, probably in 8/10 years.

References

1. Nunez, David, Malcom Smith, and Boris Bellalta. "Multi-AP Coordinated Spatial Reuse for Wi-Fi 8: Group Creation and Scheduling." arXiv preprint arXiv:2305.04846 (2023).
2. Merlin, Simone, et al. "TGax simulation scenarios." IEEE802 (2015): 11-14.