Hypercell-based energy-saving solution: architecture, protocols and AI-assisted mechanisms

The IP led the design and validation of a novel Hypercell-based energy-saving solution, defining the architecture, protocols and AI-assisted mechanisms required to dynamically switch terrestrial cells on and off while preserving user quality of service. Unlike conventional carrier shutdown approaches based on static coverage and capacity definitions and fixed thresholds, the proposed solution supports adaptive coordination between Hypercells and terrestrial cells through network-information exchange and AI-assisted building blocks.

The first task focused on the Hypercell architecture and carrier shutdown protocol [E2.12]. Two complementary architectures were investigated: a non-hierarchical solution, where the Hypercell directly coordinates terrestrial capacity cells, and a hierarchical solution in which the Hypercell coordinates legacy terrestrial coverage cells, which also act as intermediate controllers by waking up their associated terrestrial capacity cells while being able to enter sleep mode themselves. A new dynamic pairing selection algorithm was developed by evaluating several strategies, with a probabilistic fallback-cell approach providing the best performance. Based on these pairings, a distributed shutdown and wake-up protocol was designed, enabling autonomous decisions using local measurements and exchanged information.

The second task developed complementary AI-assisted building blocks intended to support future distributed Hypercell decision making. The IP supervised the development of machine learning models for traffic forecasting using NWDAF data [E1.7], AI-based SINR prediction for proactive handover quality estimation [E2.8], and the exchange of cell-level power consumption models to account for heterogeneous hardware energy profiles [E2.6] (their implementation is described in Activity 3). These components were validated independently and demonstrate how AI-driven information exchange can support traffic-aware, quality-aware and energy-aware shutdown decisions.

Overall, these activities established the main technological building blocks required for the Hypercell vision: dynamic pairing, distributed carrier shutdown and wake-up, traffic forecasting, SINR prediction and power consumption model exchange. The proposed pairing and shutdown protocol [E2.12], evaluated assuming knowledge of future traffic and homogeneous cell power consumption, achieved energy savings of up to 12.5% during low-traffic periods, while the remaining components validated complementary AI building blocks for future integrated implementations.