Advancing the integration of Artificial Intelligence to improve the efficiency and sustainability of mobile network architectures

Marco Fiore, Research Associate Professor at IMDEA Networks, highlights the research progress of the DAEMON project

03 November 2022

The DAEMON project started in January 2021 to ensure that the numerous AI algorithms that will run in network controllers and orchestrators -collectively named Network Intelligence (NI)– are systematically integrated into the mobile network architecture and can function optimally within it. There has been substantial progress towards the goal, as Marco Fiore, IMDEA Networks Research Associate Professor and the DAEMON project coordinator, stresses: “We have designed a number of original NI algorithms that have significantly advanced the state of the art in many practical network functionalities, including virtualized Radio Access Network (vRAN) scheduling, control and orchestration; Virtual Network Function (VNF) placement and control; anomaly detection, prediction and response; Edge resource and slice orchestration; anticipatory resource allocation in the network core; and Reconfigurable Intelligent Surface (RIS) control”.

The project has the potential to introduce changes in mobile networks operations that will highly benefit operators but also customers, who will enjoy better service with lower delays, higher throughput, and more resilient infrastructure. In addition, part of the efforts is dedicated to limiting the energy consumption of network infrastructures, hence reducing the footprint of mobile communications, which has a benefit for the society in these difficult times of global warming”, explains Dr. Fiore.

This European project, financed under the EU H2020-ICT-2020-2 call on Information and Communication Technology, involves 12 partners across Europe, including major industrial players in the telecommunications ecosystem, innovative SMEs, and academic institutions. Dr. Fiore remarks that “it is precisely thanks to the tight cooperation between industry and academia that the project is on track towards the planned demonstration of the viability of the NI algorithms in realistic settings, including 12 experimental testbeds, 4 simulators and data-driven tests with 14 different measurement datasets”. The consortium further proposes a first design for a novel NI plane that complements the legacy user, control and management planes that compose mobile network architectures. “The research carried out to date is definitely groundbreaking. It addresses many open problems in the automation of mobile network infrastructure operation and proposes innovative solutions based on either novel design of AI models that are tailored to network environments, or practical updates to the network architecture that improve its capability to accommodate self-management functionalities”, highlights the project coordinator.

Efficiency and sustainability

These are a few representative examples of the technical challenges addressed by the project:

  • Traditional AI algorithms take decisions to optimize a given performance metric, which in network context is provided by human experts (e.g., network managers or system engineers). However, in practical settings, the performance metric to be optimized by anticipatory network management actions is not always known a priori by the network operator. To address this challenge, DAEMON sets forth innovative NI design based on so-called loss meta-learning. This is an original concept, which lets the AI algorithm both learn how its decisions affect the performance metric and how to optimize such decisions to maximize the performance metric in very complex management tasks.
  • DAEMON is developing AI-powered solutions to attain reliability in virtualized radio access networks hosted by heavily shared resources, which reduce costs but cause fluctuations in computing capacity and hence pose a reliability hazard. The project partners address the challenge by bringing together a combination of know-how in the operation of a distributed unit (DU) and its radio processing operations and the predictive power of machine learning models.
  • The project is implementing customized AI solutions for anomaly detection in IoT platforms that depend on international mobile roaming to connect the IoT devices corresponding to different verticals (e.g., connected elevators, fleet tracking, smart meters, etc.). Given the complexity of the ecosystem that supports this global operational model that IoT verticals prefer, AI algorithms are instrumental in triggering alarms whenever IoT devices suffer from several anomalies.

The excellent progress of the research is testified by the very high scientific impact of the project results, which have been presented at all major peer-reviewed international conferences in networking, such as ACM SIGCOMM, ACM MobiCom, IEEE INFOCOM, ACM MobiSys, with over 49 papers already published by the project to date.

The technological transfer is also a strong point of DAEMON, as its consortium has currently filed 3 patents associated with solutions developed in the project, and over 25 contributions to standardization bodies such as O-RAN, ETSI, and 3GPP. One of the innovations developed in the project by Telefonica, one of the DAEMON partners, has been examined and then accepted within the European Commission’s Innovation Radar, a wide-range communication initiative supported by 23 EU Members States and aiming at fostering the most novel technological and scientific advances delivered by European innovators.

Download PDF
Source(s): IMDEA Networks Institute
Categorized in: ,

Recent Comments