D4.2–Protocol Emergence Techniques

This document summarizes the achievements made by the partners from the CENTRIC consortium, involved in the task T4.1 of the workpackage WP4, during the two and a half years of project duration. Pertaining to WP4, this report describes the most promising protocol learning techniques related to the collaborative navigation of robots, the lossless compression of the DCI messages from the control plane, 6G in-X subnetworks for industrial environments, ISAC scenarios and enhancements of the HARQ feature of MAC protocols. These findings are supported by extensive numerical simulations. In a network of a BS and several UEs, changing the flat topology to a hierarchical one yielded a three-fold averaged latency reduction. For collaborative robots, our goal-oriented multiple access scheme achieved a 10% increase in fairness of channel access time. Using Over- The-Air(OTA) lab test DCI traces, we obtained 134% gain in the average compression ratio and, in a Matlab system-level simulator, an 18% gain, over traditional compression baselines. For the sub-band allocation problem in in-X subnetworks, we obtain a 98% signalling reduction, while achieving a good control cost and block error rate trade-off. Thus, our solutions demonstrate the potential of AI/ML techniques to produce highly performant communication protocols that surpass classical ones. Their performance is due to the versatility of AI/ML algorithms to adapt to a specific task and to optimise its key performance indicators.