Periodic Reporting for period 2 - BeGREEN (Beyond 5G Artificial Intelligence Assisted Energy Efficient Open Radio Access Network)
Reporting period: 2024-01-01 to 2025-06-30
BeGREEN takes a holistic view of the radio access network to provide an evolved segment that not only accommodates increasing traffic and service levels but also considers power consumption as a factor. At a hardware level, techniques include improvement of the energy efficiency of power amplifiers and the appropriate use of accelerators to reduce power consumption relative to the use of generic compute platforms when performing network function virtualisation. At the link level, techniques to provide a better estimate of the impact of the radio channel are considered and the resulting improvement in spectral efficiency balanced against the increased power consumption associated with the resulting calculations. At a system level, new nodes and architectures are considered that move away from the assumption that additional capacity requires installation of new base-stations. These architectural considerations include enhanced capabilities and components such as Reconfigurable Intelligent Surfaces (RIS) and cell-free architectures that provide a better distribution of radio power around an area when compared to the centralised cellular approach. To achieve the expected hardware, link-level and system-level benefits, research into some fundamental capabilities are required. The use of AI/ML techniques provides a solution to reduce the number of calculations required when compared to a more traditional approach. AI/ML are also be used to recognise patterns in the system level data associated with the behaviour of the user base and to learn the most appropriate response to this behaviour in terms of both network performance and also energy consumption. The location at which these AI/ML operations are carried out within the network also have an impact on the performance of the approach, the consumption of power and the ability to share resources between different operations. The movement of data around the network to the appropriate location for calculations to be carried out also requires the definitions of new interfaces and protocols to enable the data to be processed in an open architecture of virtualised network functions. For this reason, the project will assume that the emerging O-RAN standard and its evolution is the baseline architecture. The disaggregation, virtualisation and network and service management capabilities inherent in O-RAN provide the mechanisms to realise many of the above-mentioned infrastructure changes and techniques for energy optimisation.
Determining the metrics by which power consumption should be included is a key feature of the project, and the different mechanisms by which power consumption could be reduced will be evaluated.
Determining the metrics by which power consumption should be included is a key feature of the project, and the different mechanisms by which power consumption could be reduced will be evaluated.
The work performed has focused on the identification and validation of energy-efficiency improvements across RAN elements, leading to enhanced and more sustainable radio network deployment strategies. Beyond conceptual analysis, the project has provided quantified evidence of energy savings, demonstrating that advanced architectures and control strategies can deliver substantial gains, including up to 30 dB transmit power reduction in cell-free massive MIMO, 15–84% savings through network optimisation, and up to 90% savings via relay-assisted deployments, depending on the scenario .
During Period 2, the architectural design and implementation have been the central focus. The project has advanced the baseline 3GPP and O-RAN architectures by introducing AI/ML-driven optimisation mechanisms and extending control capabilities. A key outcome is the design and validation of the BeGREEN Intelligent Plane, which enables AI-native energy optimisation across the network. This includes the definition of its components, interfaces, and integration within O-RAN, as well as the identification of architectural gaps requiring extensions (e.g. for RIS, relays, ISAC, and edge computing). The associated AI Engine, based on serverless and loosely coupled principles, provides scalable support for AI/ML services and has demonstrated up to 40% energy savings at system level without performance degradation .
At component level, WP3 has redefined the implementation of key O-RAN elements, particularly CU and DU, through optimised functional splits and hardware acceleration, targeting significant efficiency improvements beyond current SOTA deployments. For the RU, both hardware-level optimisations (e.g. PA efficiency improvements and PHY adaptations) and network-level control via RICs have been investigated to reduce energy consumption dynamically. Additionally, the integration of non-traditional components such as RIS, relays, and ISAC has been explored, demonstrating their strong potential for improving energy efficiency in diverse deployment scenarios, including challenging indoor and edge cases.
At demonstration level, under Task 5.1 multiple Proofs of Concept have been implemented and validated, confirming the feasibility and effectiveness of the proposed solutions in realistic environments. These demonstrations consolidate the project’s contributions by showing how AI-driven control, architectural innovation, and advanced radio technologies can be combined to deliver measurable energy savings while maintaining network performance.
During Period 2, the architectural design and implementation have been the central focus. The project has advanced the baseline 3GPP and O-RAN architectures by introducing AI/ML-driven optimisation mechanisms and extending control capabilities. A key outcome is the design and validation of the BeGREEN Intelligent Plane, which enables AI-native energy optimisation across the network. This includes the definition of its components, interfaces, and integration within O-RAN, as well as the identification of architectural gaps requiring extensions (e.g. for RIS, relays, ISAC, and edge computing). The associated AI Engine, based on serverless and loosely coupled principles, provides scalable support for AI/ML services and has demonstrated up to 40% energy savings at system level without performance degradation .
At component level, WP3 has redefined the implementation of key O-RAN elements, particularly CU and DU, through optimised functional splits and hardware acceleration, targeting significant efficiency improvements beyond current SOTA deployments. For the RU, both hardware-level optimisations (e.g. PA efficiency improvements and PHY adaptations) and network-level control via RICs have been investigated to reduce energy consumption dynamically. Additionally, the integration of non-traditional components such as RIS, relays, and ISAC has been explored, demonstrating their strong potential for improving energy efficiency in diverse deployment scenarios, including challenging indoor and edge cases.
At demonstration level, under Task 5.1 multiple Proofs of Concept have been implemented and validated, confirming the feasibility and effectiveness of the proposed solutions in realistic environments. These demonstrations consolidate the project’s contributions by showing how AI-driven control, architectural innovation, and advanced radio technologies can be combined to deliver measurable energy savings while maintaining network performance.
The BeGREEN project delivers results beyond the state of the art (SOTA) by combining AI-driven control, novel architectures, and validated energy-saving mechanisms with quantified performance gains. A key breakthrough is the cell-free massive MIMO architecture, which achieves up to 30 dB reduction in transmit power compared to traditional MIMO systems while maintaining performance . This demonstrates a fundamental shift toward adaptive and distributed RAN deployments.
Additional beyond-SOTA gains are achieved through network-level optimization strategies, including selective cell switch-off, RU deactivation, and traffic offloading, delivering energy savings between 15% and 84%, depending on the scenario . Furthermore, relay-assisted deployments enable up to 90% energy savings in challenging indoor environments , highlighting the impact of architectural innovation.
The introduction of the BeGREEN Intelligence Plane, integrating AI/ML into O-RAN, achieves up to 40% energy savings without performance degradation, validating the effectiveness of AI-native network control . In parallel, the project enhances reproducibility through full AI training pipelines and contributes to open-source platforms and standardization.
Overall, these results demonstrate measurable, system-level improvements that go significantly beyond current SOTA solutions in both energy efficiency and operational intelligence.
Additional beyond-SOTA gains are achieved through network-level optimization strategies, including selective cell switch-off, RU deactivation, and traffic offloading, delivering energy savings between 15% and 84%, depending on the scenario . Furthermore, relay-assisted deployments enable up to 90% energy savings in challenging indoor environments , highlighting the impact of architectural innovation.
The introduction of the BeGREEN Intelligence Plane, integrating AI/ML into O-RAN, achieves up to 40% energy savings without performance degradation, validating the effectiveness of AI-native network control . In parallel, the project enhances reproducibility through full AI training pipelines and contributes to open-source platforms and standardization.
Overall, these results demonstrate measurable, system-level improvements that go significantly beyond current SOTA solutions in both energy efficiency and operational intelligence.