6G-RAISING - 6G Radio Access Network Intelligence and Micro OrcheStratIoN for EdGe-Assisted Energy-Efficient Transceivers

6G-RAISING is proposing timely innovations that will allow the reconfiguration of key low-PHY functions -starting from the digital predistortion (DPD) one- and components included in 6G radio transceivers with the use of AI/ML techniques. To do so, key RU configuration, initialization, sensing and calibration parameters along with signal captures need to be exposed to O-RAN RICs, where AI/ML models will leverage them to apply a whole new level of automated RU reconfigurability, achieving important benefits in terms of performance and sustainable operation. Towards this end, 6GRAISING is committed to closely monitor the 3GPP and ETSI MEC standardization, along with the evolution of the O-RAN specifications.

6G-RAISING aims at proposing a complementary solution to those RL algorithms that proactively switch off base stations to reduce energy consumption at RAN level. This will be achieved by extending the concept of run-time reconfigurability throughout the entire lifespan of the RU operation by better handling the power-hungry and nonlinear by nature power amplifier, with a new Edge-assisted DPD topology. The existing RU-centric DPD paradigm will be complemented by leveraging the Edge infrastructure and RAN intelligence. Several approaches will be followed such as placing the inference ANN (DPD forward path) at the RU and the training ANN (feedback path) at Edge, or to enable Edge AI assistance to digital linearization functions, or apply runtime function replacement at FPGA level, to RU-centric DPDs. The distributed DPD processing schemes will also count on techniques able to operate with reduced sample data rates at the RU DPD observation paths. For deployments that feature similar RU characteristics continual learning options will also be evaluated within federated learning topologies. Ultimately, the idea is to expose the DPD functionality to the non-real-time RIC of O-RAN that will use AI/ML models to (predictively) reconfigure/scale/replace the DPD.