Decentralised Learning MACs for Collision-free Access in WLANs

By combining the features of CSMA and TDMA, fully decentralised WLAN MAC schemes have recently been proposed that converge to collision-free schedules. In this paper we describe a MAC with optimal long-run throughput that is almost decentralised. We then design two \changed{schemes} that are practically realisable, decentralised approximations of this optimal scheme and operate with different amounts of sensing information. We achieve this by (1) introducing learning algorithms that can substantially speed up convergence to collision free operation; (2) developing a decentralised schedule length adaptation scheme that provides long-run fair (uniform) access to the medium while maintaining collision-free access for arbitrary numbers of stations.

💡 Research Summary

The paper addresses a fundamental limitation of conventional Wi‑Fi medium access control (MAC) protocols: the trade‑off between the random, contention‑based nature of CSMA/CA and the deterministic, centrally‑scheduled nature of TDMA. While CSMA/CA suffers from escalating collision rates as the number of stations grows, TDMA requires a central scheduler and is inflexible to dynamic network sizes. Recent work has proposed hybrid, fully‑decentralized MACs that converge to collision‑free schedules, but these schemes either sacrifice throughput or require extensive coordination.

Core Contribution
The authors first formulate an “almost‑decentralized” optimal MAC model that achieves the theoretical long‑run throughput bound for any number of stations. In this model each station makes transmission decisions solely based on local observations of whether its own transmitted slot succeeded, collided, or was idle. No global time synchronization or explicit exchange of scheduling information is needed. By learning to occupy the least‑used slots, stations eventually settle into a collision‑free schedule whose length (L) satisfies (L \ge N) (where (N) is the number of active stations). The authors prove that the long‑run throughput approaches ((L-1)/L), which is arbitrarily close to the maximum possible as (L) grows.

Practical Schemes
Building on this theoretical foundation, two practically realizable schemes are introduced:

1. Fully Decentralized Learning MAC (DFS‑Full) – Each station observes three possible outcomes for every slot (success, collision, idle) and updates a probability distribution over all slots using a reinforcement‑learning rule akin to Q‑learning. The update rule is
   \