Francesc Wilhelmi

Francesc Wilhelmi

Talks and presentations

See a map of all the places I've given a talk!

Multi-armed bandits for communications

May 17, 2021

Talk, Keynote AnNet 2021, Online

Abstract: Wireless networks are evolving towards autonomous systems able to satisfy the increasing complexity behind next-generation communication standards. Artificial intelligence (AI) is expected to rule many operations held in wireless communications, including user management, accounting, security, or medium access, to name a few. In this talk, we will introduce one small piece of the AI realm, namely the multi-armed bandits (MAB) framework, which learning by experience characteristic suits many communications problems. In particular, the MAB setting allows addressing complex partial information problems by allowing an agent (or learner) to interact with the environment to accumulate knowledge and then respond to unforeseen events.

Unleashing the Potential of Machine Learning to Address Spatial Reuse in Future IEEE 802.11 WLANs: An Introduction to Two Problem Statements for the ITU AI Challenge

May 05, 2021

Talk, ITU AI/ML in 5G Challenge, Online

Abstract: Spatial Reuse (SR) is one of the techniques that is gaining more attention in next-generation IEEE 802.11 standards. SR was firstly introduced in the IEEE 802.11ax (11ax) as a decentralized mechanism, but it is now evolving with IEEE 802.11be (11be) thanks to the multi Access Point (multi-AP) feature. In both cases, SR aims at increasing the number of parallel transmissions in an Overlapping Basic Service Set (OBSS) by applying sensitivity adjustment and transmit power control. In this talk, we will provide an overview of both 11ax and 11be SR mechanisms and discuss two problem statements for the ITU AI Challenge, where participants must harness the potential of Machine Learning (ML) to solve a relevant problem in communications such as the SR one.

Towards Spatial Reuse in Future WLANs: a Sequential Learning Approach

October 07, 2020

Talk, Thesis defence - Universitat Pompeu Fabra, Online (Spain)

Abstract: The Spatial Reuse (SR) operation is gaining momentum in the latest IEEE 802.11 family of standards due to the overwhelming requirements posed by next-generation wireless networks. In particular, the rising traffic requirements and the number of concurrent devices compromise the efficiency of increasingly crowded Wireless Local Area Networks (WLANs) and throw into question their decentralized nature. The SR operation, initially introduced by the IEEE 802.11ax-2021 amendment and further studied in IEEE 802.11be-2024, aims to increase the number of concurrent transmissions in an Overlapping Basic Service Set (OBSS) using sensitivity adjustment and transmit power control, thus improving spectral efficiency. Our analysis of the SR operation shows outstanding potential in improving the number of concurrent transmissions in crowded deployments, which contributed to enabling low-latency next-generation applications. However, the potential gains of SR are currently limited by the rigidity of the mechanism introduced for the 11ax, and the lack of coordination among BSSs implementing it. The SR operation is evolving towards coordinated schemes where different BSSs cooperate. Nevertheless, coordination entails communication and synchronization overhead, which impact on the performance of WLANs remains unknown. Moreover, the coordinated approach is incompatible with devices using previous IEEE 802.11 versions, potentially leading to degrading the performance of legacy networks. For those reasons, in this thesis, we start assessing the viability of decentralized SR, and thoroughly examine the main impediments and shortcomings that may result from it. We aim to shed light on the future shape of WLANs concerning SR optimization and whether their decentralized nature should be kept, or it is preferable to evolve towards coordinated and centralized deployments. To address the SR problem in a decentralized manner, we focus on Artificial Intelligence (AI) and propose using a class of sequential learning-based methods, referred to as Multi-Armed Bandits (MABs). The MAB framework suits the SR problem because it addresses the uncertainty caused by the concurrent operation of multiple devices (i.e., multi-player setting) and the lack of information in decentralized deployments. MABs can potentially overcome the complexity of the spatial interactions that result from devices modifying their sensitivity and transmit power. In this regard, our results indicate significant performance gains (up to 100% throughput improvement) in highly dense WLAN deployments. Nevertheless, the multi-agent setting raises several concerns that may compromise network devices’ performance (definition of joint goals, time-horizon convergence, scalability aspects, or non-stationarity). Besides, our analysis of multi-agent SR encompasses an in-depth study of infrastructure aspects for next-generation AI-enabled networking.

ITU AI/ML in 5G Challenge: Improving the capacity of IEEE 802.11 WLANs through Machine Learning

October 07, 2020

Talk, ITU AI/ML in 5G Challenge, Online

Abstract: The talk is intended to present, describe and solve doubts concerning the problem statement “Improving the capacity of IEEE 802.11 WLANs through Machine Learning” https://www.upf.edu/web/wnrg/ai_challenge of the ITU AI/ML in 5G Challenge. The first part of the talk will be devoted to introducing IEEE 802.11 WLANs and the Channel Bonding (CB) problem. Then, the dataset will be described together with the goals of the challenge.

The ITU-T AI Challenge at the UPF: Improving the capacity of IEEE 802.11 WLANs through Machine Learning

July 15, 2020

Talk, Foro internacional sobre nuevas tecnologías de la información y comunicación - Universidad Técnica del Norte, Online (Ecuador)

Abstract: The talk is intended to present, describe and solve doubts concerning the problem statement “Improving the capacity of IEEE 802.11 WLANs through Machine Learning” (https://www.upf.edu/web/wnrg/ai_challenge) of the ITU AI/ML in 5G Challenge. The first part of the talk will be devoted to introducing IEEE 802.11 WLANs and the Channel Bonding (CB) problem. Then, the dataset will be described together with the goals of the challenge.

[ML5G-I-238] Machine Learning Sandbox for future networks including IMT-2020: requirements and architecture framework

June 02, 2020

Contribution, 9th Meeting of the FG-ML5G, ITU-T, Online

Abstract: Use cases for integrating machine learning (ML) to future networks including IMT-2020 has been documented in Supplement 55 and an architecture framework for this integration was specified in ITU-T Y.3172. However, network stakeholders are apprehensive about using ML-driven approaches directly in live networking systems because it can lead to unexpected situations that can degrade KPIs. This is mostly due to the apparent complexity of ML mechanisms (e.g., deep learning), the incompleteness of the available training data, the uncertainty produced by exploration-exploitation approaches (e.g., reinforcement learning), etc. In the face of such impediments, the ML Sandbox emerges as a potential solution that allows mobile network operators (MNOs) for improving the degree of confidence in ML solutions before their application to the network infrastructure. This contribution discusses the requirements, architecture, and implementation examples for ML Sandbox in future networks including IMT-2020.

[ML5G-I-176] Adoption of the ITU-T’s Architecture in IEEE 802.11 WLANs

November 05, 2019

Contribution, 7th Meeting of the FG-ML5G, ITU-T, Berlin, Germany

Abstract: A lot of hopes have been placed in Machine Learning (ML) as a key enabler of future wireless networks. By taking advantage of the large volumes of data generated by networks, ML is expected to deal with the ever-increasing complexity of networking problems, ranging from cloud-based (centralized) to edge-computing-like approaches (decentralized). Unfortunately, current networking systems are not yet prepared for supporting the ensuing requirements of ML-based applications, especially for enabling procedures related to data collection, processing, and output distribution. This paper addresses the challenges of re-designing the network operation to accommodate the utilization of ML into future Wireless Local Area Networks (WLANs). For that, we propose to adopt the International Telecommunications Union (ITU) unified architecture for future networks. Based on the adoption of this architecture, insights are provided into the major challenges of introducing ML to WLANs. Moreover, we describe a set of use cases in which ML is applied to WLAN-related problems.

Towards the implementation of 11ax features in Komondor

June 21, 2019

Conference, Workshop on Next-Generation Wireless with ns-3, Florence, Italy

Abstract: Presentation of the Komondor simulator, where novel 11ax-based features are implemented. Particular emphasis was put on the Spatial Reuse (SR) part.

A proposal to reuse Komondor, an IEEE 802.11ax-oriented simulator for future wireless networks in FG ML5G

March 07, 2019

Contribution, 5th Meeting of the FG-ML5G, ITU-T, Shenzhen, China

Abstract: Komondor is an open-source IEEE 802.11ax-oriented simulator that includes intelligent agents as an integral part of the wireless operation. As part of this simulator, a common ML-based architecture is proposed to frame different types of learning, which range from decentralized to centralized systems, according to the available information. We propose to use Komondor to contribute to the FG-ML5G in a twofold manner: i) serve as a test environment for several purposes of the group (implementation of proposals, representation of data formats, etc.), and ii) generate synthetic data as a result of the IEEE 802.11 operation in a customized way. This document is intended as a starting point to showcase the potential of the simulator through specific use cases, so that interactions and potential collaboration with the FG can be found.

Implications of Decentralized Learning in Dense WLANs - Submission to FG-ML5G First Meeting

January 30, 2018

Contribution, 1st Meeting of the FG-ML5G, ITU-T, Geneva, Switzerland

Abstract: Understanding the consequences of applying Reinforcement Learning (RL) in dense and uncoordinated environments (e.g., Wi-Fi) is critical to optimize the performance of next-generation wireless networks. In this document we present a decentralized approach in which Wireless Networks (WNs) attempt to learn the best possible configuration in an adversarial environment according to their own performance. In particular, we provide a Multi-Armed Bandits (MABs) based model in which devices are allowed to tune their frequency channel, transmit power and Carrier Sense Threshold (CST). Our results show that, despite using only local information, a collaborative behavior can be obtained among independent devices that share the same resources. Furthermore, we study the effects of applying such method under different equilibrium situations with respect to the adversarial setting. Finally, some insights are provided regarding the consequences of applying learning in presence of legacy nodes.

Improving Spatial Reuse in High-Density Wireless Networks through Learning

March 01, 2017

Poster, 5th EITIC Doctoral Student Workshop, Barcelona, Spain

Abstract: Popularity of IEEE 802.11 Wireless Local Area Networks (WLANs) leads to massive deployments in which few frequency resources must be shared, resulting in inefficiency in dense environments. The behavior of the protocols that grant access to the medium, which are based on Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA), is inefficient in dense deployments, and prone to generate issues such as the Hidden-Terminal and the Exposed-Terminal problems. Therefore, the overall throughput may be considerably reduced due to collisions and/or starvation. As the complexity of Wireless Networks in terms of variability prevents to computationally find the optimal configuration for a given network, we aim to use Reinforcement Learning (RL) to find close-to-optimal solutions adaptively.