D5.3 BeGREEN Use Case Demonstration and KPI Assessment

BeGREEN D5.3 documents the consortium’s final system-level demonstrations and quantitative KPI assessment after thirty months of R&D. Five Proofs-of-Concept (PoCs) were built, at BT Adastral Park (UK), Accelleran, IHP, NEC Labs, and RunEL premises, and partner universities, showcasing a holistic, AI-driven, energy-efficient 6G Radio Access Network (RAN) architecture. Key results reported in this deliverable are: · PoC1: Intelligence-Plane maturity ‑ Non-RT and Near-RT RICs, AI Engine and rApp/xApp suite reached TRL-6 and executed a closed-loop cell on/off strategy that reduced radio-access energy by up to 30 % with <15 % Quality of Service (QoS) impact. A conflict manager ensured >30% energy-efficiency gain while safeguarding mobility-related xApps, proving the feasibility of multi-objective orchestration. Model-Selection logic trimmed Machine Learning (ML)-inference energy by 40–68% while maintaining accuracy through adaptive time-granularity. · PoC2: Reconfigurable Intelligent Surface (RIS)-assisted Integrated Sensing and Communication (ISAC) ‑ Sub-6 GHz sensing achieved 0.7 m range resolution and >90% person-count precision. First-of-its-kind demonstration of “sensing-through-RIS” verified user detection behind obstacles, enabling coverage-aware RU sleep modes. · PoC3: Energy-aware CU & O-RAN RIC ‑ Porting to ARM servers yielded up to 10% savings under high traffic, with 33% lower idle power versus x86, validating edge-grade, low-carbon cloud-native RAN. · PoC4: Hardware (HW)-accelerated DU - Graphics Processing Unit (GPU)-based sphere decoder cut execution time by 11× and improved energy efficiency from 9 kbit/J (CPU) to 100 kbit/J—evidence that off-the-shelf accelerators are pivotal for green Layer-1 processing. · PoC5: Power Amplifier (PA)-blanking in 6G RU ‑ Field trial at Brunel University confirmed 64 % PA power savings, dropping RU PA draw from 5.41 W to 1.94 W using scheduler-driven blanking. Collectively, the PoCs demonstrate an end-to-end 6G blueprint capable of achieving 20-35 % system-level energy reduction without degrading user experience, supporting the project’s vision of a carbon-aware mobile network stack. The results are fed into O-RAN Alliance and 3GPP work items.