Page 83 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 4 – AI and machine learning solutions in 5G and future networks

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 4

MACHINE LEARNING FOR PERFORMANCE PREDICTION OF CHANNEL BONDING IN
NEXT‑GENERATION IEEE 802.11 WLANS

4
4
2
3
1
1
Francesc Wilhelmi , David Góez , Paola Soto , Ramon Vallés , Mohammad Alfai i , Abdulrahman Algunayah ,
6
5
5
6
Jorge Martı́n‑Pérez , Luigi Girletti , Rajasekar Mohan , K Venkat Ramnan , Boris Bellalta 1
2
3
1 Universitat Pompeu Fabra, Spain, Universidad de Antioquia, Colombia, University of Antwerp, Belgium,
5
6
4 Saudi Telecom, Saudi Arabia, Universidad Carlos III de Madrid, Spain, PES University, India
NOTE: Corresponding author: Francesc Wilhelmi, fwilhelmi@cttc.cat
Abstract – With the advent of Arti icial Intelligence (AI)‑empowered communications, industry, academia, and standard‑
ization organizations are progressing on the de inition of mechanisms and procedures to address the increasing complexity
of future 5G and beyond communications. In this context, the International Telecommunication Union (ITU) organized the
irst AI for 5G Challenge to bring industry and academia together to introduce and solve representative problems related to
the application of Machine Learning (ML) to networks. In this paper, we present the results gathered from Problem State‑
ment 13 (PS‑013), organized by Universitat Pompeu Fabra (UPF), whose primary goal was predicting the performance of
next‑generation Wireless Local Area Networks (WLANs) applying Channel Bonding (CB) techniques. In particular, we pro‑
vide an overview of the ML models proposed by participants (including arti icial neural networks, graph neural networks,
random forest regression, and gradient boosting) and analyze their performance on an open data set generated using the
IEEE 802.11ax‑oriented Komondor network simulator. The accuracy achieved by the proposed methods demonstrates the
suitability of ML for predicting the performance of WLANs. Moreover, we discuss the importance of abstracting WLAN inter‑
actions to achieve better results, and we argue that there is certainly room for improvement in throughput prediction through
ML.
Keywords – Channel bonding, IEEE 802.11 WLAN, ITU Challenge, machine learning, network simulator

1. INTRODUCTION which allows deriving models from experience. Never‑
theless, the adoption of AI/ML in networks is still in its
The utilization of Arti icial Intelligence (AI) and Machine initial phase, and a lot of work needs to be done. In
Learning (ML) techniques is gaining momentum to ad‑ this regard, standardization organizations are undertak‑
dress the challenges posed by next‑generation wireless ing signi icant efforts towards fully intelligent networks.
communications. The fact is that networks are nowa‑ An outstanding example can be found in the Interna‑
days facing unprecedented levels of complexity due to tional Telecommunication Union (ITU)’s ML‑aware archi‑
novel use cases including features such as spatial multi‑ tecture [1], which lays the foundations of pervasive ML
plexing, multi‑array antenna technologies, or millimeter for networks.
wave (mmWave) communications. While these features
allow providing the promised performance requirements Another aspect essential for the prosperity of AI/ML in
in terms of data rate, latency, or energy ef iciency, their communications is data availability and openness. In this
implementation entails additional complexity (especially context, the ITU AI/ML in 5G Challenge [2] was set in mo‑
for crowded and highly dynamic deployments), thus mak‑ tion to encourage industry, academia, and other stake‑
ing hand‑crafted solutions unfeasible. holders to collaborate and exchange data for solving rel‑
evant problems in the ield. This initiative entailed a big
IEEE 802.11 Wireless Local Area Networks (WLANs) are step forward in bringing open source closer to standards.
one of the most popular access solutions in the unlicensed
band, and they represent a prominent example of increas‑ As a contribution to the ITU challenge, and aligned with
ing complexity in wireless networks. The optimization WLANs optimization, in this paper, we present the re‑
of WLANs underlines particular challenges due to the sults obtained from problem statement “Improving the
capacity of IEEE 802.11 Wireless Local Area Networks
decentralized nature of these types of networks, which
mostly operate using Listen‑Before Talk (LBT) transmis‑ (WLANs) through Machine Learning” (referred to as PS‑
sion procedures. If to this we add that WLAN deploy‑ 013 in the context of the challenge), whereby participants
ments are typically unplanned, dense, and highly dy‑ were called to design ML models to predict the perfor‑
namic, the complexity is even increased. mance of next‑generation Wi‑Fi deployments. In this
article, we gather all the work done in the context of
To address the optimization of next-generation WLANs, the above‑mentioned problem statement and provide
the usage of AI/ML emerges as a compelling solution by a compilation of the proposed ML models used
leveraging useful information obtained across data, to address the problem of CB in WLANs.

© International Telecommunication Union, 2021 67