Talks
Enhancing 5G Networks with AI: A Journey from Data Analytics to Large Language Models
Advanced Wireless Communications Seminars
February 5th, 2024
Telecom SudParis, Institut Polytechnique de Paris, Paris, France
5G-Energy Consumption Modelling: AI/ML solutions for Climate Change
ITU AI for Good
October 27th, 2023
Online
Abstract
Link
Video
As we advance toward the global adoption of 5G networks, it becomes essential not only to boost their speed but also enhance their sustainability. The surge in the number of connected devices and the data explosion they induce has imposed a significant challenge on network operators: how to meet the escalating demand without skyrocketing the energy consumption.
Within this context, the energy optimization of base stations emerges as a pivotal aspect of constructing a sustainable and cost-efficient 5G network. Given the myriad factors that impact this energy consumption, having precise models that elucidate how different configurations and parameters affect energy usage is of utmost importance.
September 14th, 2023
Universitat Politecnica de Valencia, Valencia, Spain
Abstract
The fifth generation (5G) of radio technology is revolutionizing our everyday lives, by enabling a high degree of automation, through its larger capacity, massive connectivity, and ultra-reliable low-latency communications. However, while 5G networks offer unprecedented capabilities, there exists a crucial area that demands further enhancement: energy efficiency. Despite achieving around a 4x improvement in energy efficiency compared to 3GPP Long Term Evolution (LTE) deployments, current 5G New Radio (NR) networks still consume up to 3x more energy, leading to heightened carbon emissions and increased operational costs for network operators.
In light of the increasing interest in this field, this talk confronts the urgent challenge of 5G energy efficiency from an industrial perspective. We embark on a journey that harnesses the potential of big data and machine learning, exploring how data gleaned from thousands of base stations can be used to construct precise machine learning models to characterize the energy consumption of multi-carrier base stations. The insights gained from this framework is leveraged to create a realistic and analytically tractable power consumption model, which can drive theoretical analyses, standardization, development, and optimization frameworks. Moreover, we peer into the horizon of a future where large language models take center stage, autonomously generating explainable models, thus advancing the pursuit of energy-efficient networks.
July 11th, 2023
Online
Abstract
Link
Video
As we advance toward the global adoption of 5G networks, it becomes essential not only to boost their speed but also enhance their sustainability. The surge in the number of connected devices and the data explosion they induce has imposed a significant challenge on network operators: how to meet the escalating demand without skyrocketing the energy consumption.
Within this context, the energy optimization of base stations emerges as a pivotal aspect of constructing a sustainable and cost-efficient 5G network. Given the myriad factors that impact this energy consumption, having precise models that elucidate how different configurations and parameters affect energy usage is of utmost importance.
This is the juncture at which machine learning enters the picture. It potentially offers a robust tool for developing meticulous energy models and optimizing the energy consumption of the network. This challenge is designed to tackle this crucial concern, enabling participants to contribute towards crafting more energy-efficient 5G networks
February 14th, 2023
Telecom SudParis, Institut Polytechnique de Paris, Paris, France
Machine Learning and Analytical Power Consumption Models for 5G Base Stations
Next Generation Mobile Networks (NGMN) Green Future Networks
December 2nd, 2021
Online
Forecasting Mobile Traffic to Achieve Greener 5G Networks: When Machine Learning is Key
Huawei AI forum
May 27th, 2021
Huawei, Paris, France
European Training Networks: My experience
WindMill kickoff school
November 19th, 2019
Aalborg University, Aalborg, Denmark
Energy-aware network control
SoftCom 2019
September 22th, 2019
Split, Croatia
Abstract
The talk focuses on the control methods for the management of mobile networks with energy harvesting capabilities. We introduce a two-tier mobile network consisting of macro base stations and renewable energy powered small base stations. Online optimization approaches based on machine learning are proposed as a way to optimize the utilization of the harvested energy by intelligently switching on/off the small base stations. Moreover, Multi-access Edge Computing (MEC) is introduced, where the dynamic placement of functional splits is adopted as a way to further reduce the mobile network energy consumptions.
September 27th, 2018
CTTC, Barcelona, Spain
Network resource allocation policies with energy transfer capabilities
1st CTTC Workshop
September 21th, 2018
Sitges, Spain
Abstract
This project investigates on possible integration architectures between the energy harvesting mobile network and the smart
electricity grid. The main scope is to study the capability of 5G mobile networks of intelligently routing energy in a micro grid
of interconnected conventional/renewable energy sources and loads, in order to (i) satisfy the demand of communication networks
while avoiding energy outages in zones with high user density and/or low ambient energy availability and (ii) to provide ancillary
services to the grid.
November 16th, 2017
CTTC, Barcelona, Spain
Abstract
In this talk we introduce our work in SCAVENGE, an European Training Network. The project is about sustainable design for mobile communication systems by engineering the integration and the control of renewable energy sources within communication network elements, such as base stations and mobile phones. We will describe the main objectives of the Network, detailing the training and research activities that we have been carrying out during our first year of work.
September 21th, 2017
Paris, France
Abstract
This project investigates on possible integration architectures between the energy harvesting mobile network and the smart electricity grid. The main scope is to study the capability of 5G mobile networks of intelligently routing energy in a micro grid
of interconnected conventional/renewable energy sources and loads, in order to (i) satisfy the demand of communication networks
while avoiding energy outages in zones with high user density and/or low ambient energy availability and (ii) to provide ancillary
services to the grid.
June 2nd, 2017
Strathclyde University, Glasgow, UK
Abstract
Current trends anticipate that 5G mobile networks will be composed of ultra-dense deployments of heterogeneous Base Stations (BSs), where BSs using different transmission powers coexist to provide the 1000x network capacity increase that is required by 2020. Accordingly, the traditional macro cell layer will be complemented or replaced with multiple overlapping tiers of smaller cells, which extend the system capacity, thanks to a higher spatial reuse and to a better spectral efficiency. We introduce a two-tier cellular network architecture, where self-sustainable small cells (SCs), solely relying on energy harvesting and storage, can offload the grid-connected macro base station. The available energy at the SC must be allocated in an optimal way, in order to jointly minimize the consumption of grid energy and maximize the system performance. The proposed offline optimal algorithm, based on the label correcting algorithm, finds the optimal energy allocation policy, after transforming the problem into a shortest-path problem.
November 21st, 2016.
CTTC, Barcelona, Spain
Data analytics for Smart Parking Applications
DEI Awards June 5th, 2015.
University of Padova, Padova, Italy
Tutorials
Data-Driven Modelling and Optimization of Green Future Mobile Networks: From Machine Learning to Generative AI
N. Piovesan, A. De Domenico, D. López-Pérez
IEEE International Conference on Machine Learning for Communications and Networking, Stockholm (Sweden), 5 May 2024
Abstract
Conference
The fifth generation (5G) of radio technology is revolutionizing our everyday lives, by enabling a high degree of automation, through its larger capacity, massive connectivity, and ultra-reliable low-latency communications. Moreover, 5G technology is allowing for the first time to expand cellular systems into new ecosystems, thus impacting every industry. Despite its unprecedented capabilities, however, 5G networks can —and must— further improve in certain key technology areas, such as that of energy efficiency. While current third generation partnership project (3GPP) new radio (NR) deployments provide an improved energy efficiency of around 4x w.r.t. 3GPP long term evolution (LTE) ones, they still consume up to 3x more energy. This is mostly due to the more processing required to handle the wider bandwidth and the more antennas, and is resulting in increased carbon emissions and electricity bills for operators. Even if the 3GPP NR specification provides a rich set of tools to meet IMT-2020 energy efficiency requirements, such as carrier, channel, and symbol shutdown, among others, it is important to note that one of the main energy consumption challenges of 5G networks is the complexity of their optimization in wide-area deployments: A large-scale, stochastic, non-convex and non-linear optimization problem. In light of the increasing interest in this field, this one-of-a-kind tutorial shares the author's industrial and academic views on this 5G energy efficiency problem. In more details, the tutorial provides a fresh look at energy efficiency enabling technologies in 3GPP NR, in particular, massive MIMO, carrier aggregation, the lean carrier design and different shutdown methods. By leveraging on the concepts of big data and machine learning, the tutorial presents practical scenarios in which data collected from thousands of base stations can be successfully used to derive accurate machine learning models for the main building blocks of the energy efficiency optimization problem. Furthermore, it explores the possibility of a future where generative AI plays a central role in autonomously generating explainable models, thus advancing the quest for energy-efficient networks. In this context, the tutorial delves into the adoption of large language models (LLMs) in the industry and the need for foundational telecom models and specific evaluation frameworks for generative AI.
Theodore S. Rappaport,
N. Piovesan
IEEE Wireless Communications and Networking Conference (WCNC), Dubai (UAE), 21 April 2024
Abstract
Conference
As the telecommunications industry continues to roll out 5G and begins to shift towards 6G technology, a significant challenge is emerging: the substantial increase in power consumption. This situation highlights the urgent need for more energy-efficient technologies. Currently, although various metrics are used to evaluate the energy efficiency of specific parts of a network, there is a clear need for a universal standard that can accurately assess the energy consumption of telecommunications systems. Although various metrics are available today to assess the energy efficiency of different network elements, the absence of a universal metric tailored for evaluating telecommunication systems is evident.
Drawing inspiration from Harold Friis's seminal work on Noise Factor in 1944, this tutorial introduces the Waste Factor as a novel metric designed for assessing power waste in telecommunication systems. To demonstrate that the Waste Factor is a universal metric, we apply it to various scenarios. These include 5G and 6G Radio Access Networks (RANs) and Data Centers. This approach highlights the use of Waste Factor to assess the energy efficiency of both individual network components and the overall system. In these scenarios, we present a detailed comparison between prevalent energy efficiency metrics and Waste Factor, highlighting its significance for both current and future network technologies. Importantly, our tutorial includes an in-depth analysis of RAN site architectures, wherein we assess the Waste Factor of key RAN components and configurations based on real measurements. Based on this analysis, we assess the influence of multiple deployment parameters and energy-saving methods on the energy-throughput trade-off in 5G and 6G networks. Finally, we highlight that the Waste Factor is key to providing both comprehensive theoretical analysis and support system design optimization, which will pave the way for greener and more sustainable communication systems.
, A. De Domenico, D. López-Pérez
European Conference on Networks and Communications (EuCNC) & 6G Summit , 6-9 June 2023, Gothenburg (Sweden)
Abstract
Conference
The fifth generation (5G) of radio technology is changing our everyday lives, by enabling a plethora of new use cases, through its better coverage, larger capacity and massive connectivity. Thanks to its ultra-reliable low-latency communications, 5G also allows a high degree of automation, thus helping to expand cellular systems into new ecosystems. Importantly, 5G has already become an integral part of governmental and industrial environmental programs, as it is envisioned that an intelligent exploitation of their resources through 5G will significantly decrease carbon emissions. Despite its unprecedented capabilities, however, 5G networks must further improve in certain key technology areas, particularly in that of network energy efficiency. While current third generation partnership project (3GPP) new radio (NR) deployments provide an improved energy efficiency of around 4x w.r.t. 3GPP long term evolution (LTE) ones, they still consume up to 3x more energy, resulting in increased carbon emissions and electricity bills for operators. Even if the 3GPP NR specification provides a rich set of tools to
meet IMT-2020 energy efficiency requirements, it is important to note that one of the main challenges to 5G network energy efficiency is the complexity of their optimization in wide-area deployments: a large-scale, stochastic, non-convex and non-linear optimization problem. In light of the increasing interest in this field, this one-of-a-kind tutorial shares the author’s industrial view on the 5G energy efficiency challenge. This tutorial provides a detailed, up-to-date overview of the most relevant technologies that a 5G radio access network can use to increase its energy efficiency from both a theoretical and practical perspective. Moreover, this tutorial shows how increasing the network energy efficiency by exploiting such technologies in practical scenarios highly depends on the accuracy of the models used to characterize the network. In this line, this tutorial exhaustively surveys and presents machine learning techniques which are being used to create accurate network models for most network components and processes, and optimize a large-scale 5G network.
, A. De Domenico, D. López-Pérez
IEEE International Conference on Communications (ICC), 16-20 May 2022, Seoul (South Korea)
Abstract
Conference
The fifth generation (5G) of radio technology is revolutionizing our everyday lives, by enabling a high degree of automation, through its larger capacity, massive connectivity, and ultra-reliable low-latency communications. Despite its capabilities, however, 5G networks must improve in certain key technology areas, such as that of energy efficiency. While current 3GPP NR deployments provide an improved energy efficiency of around 4x w.r.t. 3GPP LTE ones, they still consume up to 3x more energy. Even if the 3GPP NR specification provides several tools to meet IMT-2020 energy efficiency requirements, one of the main energy consumption challenges of 5G networks is the complexity of their optimization in wide-area deployments: A large-scale, stochastic, non-convex and non-linear optimization problem. In light of the increasing interest in this field, this one-of-a-kind tutorial shares the author's industrial view on this 5G energy efficiency problem. In details, the tutorial provides a fresh look at energy efficiency enabling technologies in 3GPP NR. Moreover, by leveraging on the concepts of big data and machine learning, the tutorial presents practical scenarios in which data collected from thousands of base stations can be used to derive accurate machine learning models for the main building blocks of the energy efficiency optimization problem.
Panels
The interplay of Large Generative Models and Future Networks
A. De Domenico,
N. Piovesan, D. López-Pérez
IEEE International Conference on Machine Learning for Communications and Networking, Stockholm (Sweden), 6 May 2024
