Szczegóły publikacji
Opis bibliograficzny
Learning-based channel access in Wi-Fi: a multi-armed bandit approach / Miguel Casasnovas, Francesc Wilhelmi, Richard Combes, Maksymilian WOJNAR, Katarzyna KOSEK-SZOTT, Szymon SZOTT, Anders Jonsson, Luis Esteve-Elfau, Boris Bellalta // IEEE Access [Dokument elektroniczny]. — Czasopismo elektroniczne ; ISSN 2169-3536 . — 2026 — vol. 14, s. 125904-125922. — Wymagania systemowe: Adobe Reader. — Bibliogr. s. 125921-125922, Abstr. — Publikacja dostępna online od: 2026-08-07
Autorzy (9)
- Casasnovas Miguel
- Wilhelmi Francesc
- Combes Richard
- AGHWojnar Maksymilian
- AGHKosek-Szott Katarzyna
- AGHSzott Szymon
- Jonsson Anders
- Esteve-Elfau Luis
- Bellalta Boris
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 169880 |
|---|---|
| Data dodania do BaDAP | 2026-09-08 |
| Tekst źródłowy | URL |
| DOI | 10.1109/ACCESS.2026.3721756 |
| Rok publikacji | 2026 |
| Typ publikacji | artykuł w czasopiśmie |
| Otwarty dostęp |  |
| Creative Commons | |
| Czasopismo/seria | IEEE Access |
Abstract
Due to largely static protocol configurations, IEEE 802.11 (Wi-Fi) channel access offers limited adaptability to dynamic network conditions, particularly in dense and overlapping deployments, leading to inefficient spectrum utilization, increased contention, and packet collisions. This paper investigates reinforcement learning (RL) as a data-driven, decentralized, and online approach for adaptive Wi-Fi medium access control (MAC). In particular, we consider multi-armed bandit (MAB) strategies for the joint selection of the primary channel, channel width, and contention window (CW). In this setting, we systematically study key design choices, including the adoption of joint action spaces, where a single agent (SA) optimizes all parameters, or factorized action spaces, where multiple agents (MA) handle each parameter independently, as well as the incorporation of contextual information, evaluated through contextual and non-contextual MAB formulations. Simulation results provide insights into these design choices, showing that MA architectures converge faster than their SA counterparts due to smaller action spaces, and that using contextual information consistently improves performance in the considered scenarios. In multi-player settings, decentralized learners achieve implicit coordination; however, their greedy behavior may degrade the performance of coexisting networks and induce policy-chasing dynamics. Overall, our findings suggest that (contextual) MAB-based learning provides a lightweight and adaptive alternative to static IEEE 802.11 mechanisms, enabling more efficient and intelligent spectrum utilization, and provides practical insights for learning-driven, MAB-based MAC protocols.
