Machine Learning for Wireless Networking in Highly Dynamic Scenarios

Introduction

The interest on using efficiently the Industrial, Scientific and Medical (ISM) spectrum bands for wireless communications is growing also among telecommunications operators offering broadband Internet access. However, due to the free-of-use conditions in ISM bands, and given their current and expected future congestion levels, offering service guarantees to the final users is a very challenging mission due to the complex and highly dynamic interactions between networks competing for the same spectrum resources.

In a network environment of such a complexity, heterogeneity and unpredictability, the current resource management strategies are not capable enough to keep the system working under optimal conditions. In addition, since networks will rarely be in steady state, the use of traditional techniques to analyze the system performance is unfeasible. Under these conditions, there is a growing interest in the use of machine learning techniques, capable of understanding the interactions between competing networks, exploring the effect of different actions, and intelligently choosing those that maximize the overall performance of the system.

Despite some recent interest by the research community and industry, the use of machine learning techniques for resource management in wireless networks is still in its infancy. One of the first questions to solve is where to implement those functionalities to maximize their potential. In the one hand, doing it at the cloud guarantees high computing power and large data availability, thus allowing to take accurate decisions about future network states, but at the cost of a large response delay. In the other hand, moving it to the edge limits both the available data and computational power, but allows to respond fast to a changing environment.

In that crossroad, this project aims to combine efficiently both approaches, i.e., short-term local decisions (based on partial information) with much more accurate but mid/long decisions (based on a global view of the network). This is still and unexplored research path with many challenges in front, including the design of compatible machine learning methods, the definition of the action domains between decision points, and a qualitative and quantitative performance evaluation that takes into account the actions costs and potential overheads, among others aspects.

Solving those challenges is the main goal of this project. We will design and evaluate a network architecture, new functionalities and protocols able to efficiently combine both decentralized and centralized optimization decisions based on the use of machine learning and data analytics. We will also characterize the interactions between local and global decision points (agents), studying both transient and stationary regimes (if exist), stability conditions, and their ability and speed to convergence to optimal solutions.

We expect the project will have a high impact in the research community, standardization committees, and industrial sector, providing both a theoretical and practical framework to test the proposed approach, new functionalities, and machine learning algorithms for key future dense WLAN scenarios. Using our approach we aim to significantly improve the spectrum utilization efficiency of ISM bands, with the goal to achieve a x10 gain with respect to the Quality of Experience observed by final users nowadays.

 

Publications

Journals

Álvaro López Raventós, Boris Bellalta; "Concurrent Decentralized Channel Allocation and Access Point Selection using Multi-Armed Bandits in multi BSS WLANs", Computer Networks, 2020.

F. Wilhelmi, S. Barrachina-Muñoz, B. Bellalta, C. Cano, A. Jonsson, V. Raml.; “A Flexible Machine Learning-Aware Architecture for Future WLANs”. IEEE Wireless Magazine 2020

Marc Carrascosa-Zamacois, Boris Bellalta; "Multi-armed bandits for decentralized AP selection in enterprise WLANs". Computer Communications, 2020.

S. Barrachina, F. Wilhelmi, B. Bellalta; "Online Primary Channel Selection for Dynamic Channel Bonding in High-Density WLANs". IEEE Wireless Communication Letters, 2020.

 

Conferences / Workshops

S. Barrachina, B. Bellalta, E. Knightly ; "Wi-Fi All-Channel Analyzer". ACM WinTech 2020 (a Mobicom 2020 workshop).

Dovelos, K., Matthaiou, M., Ngo, H. Q., & B. Bellalta; “Massive MIMO with Multi-Antenna Users under Jointly Correlated Ricean Fading”. IEEE International Conference on Communications (ICC), 2020.

Francesc Wilhelmi, Sergio Barrachina, B. Bellalta; "On the Performance of the Spatial Reuse Operation in IEEE 802.11ax WLANs". IEEE Conference on Standards for Communications and Networking (CSCN) 2019.

Francesc Wilhelmi and Sergio Barrachina-Muñoz; "Towards the implementation of 11ax features in Komondor". Workshop on Next-Generation Wireless with NS-3 (WNGW), 2019. [Invited talk]

Alvaro Lopez, Francesc Wilhelmi, Sergio Barrachina, B. Bellalta; "Combining Software Defined Networks and Machine Learning to enable Self Organizing WLANs". Sixth International Workshop on Cooperative Wireless Networks - 2019 (CWN'19) / WIMOB 2019, 2019.

Carrascosa, Marc, and Boris Bellalta. "Decentralized AP selection using Multi-Armed Bandits: Opportunistic ε-Greedy with Stickiness." In 2019 IEEE Symposium on Computers and Communications (ISCC), pp. 1-7. IEEE, 2019.