Intelligent 5g l2 mac scheduler

Project Marconi includes a series of innovations targeted towards realizing cognitive capabilities as part of Intelligent RAN Controller nodes. It utilizes Capgemini 5G ORAN RIC platform named RATIO along with Intel Flex RAN 5G Layer 1 implementation. The key idea is to enhance network performance in terms of spectral effi ciency, quality of service (QoS), and network resource utilization. The Marconi team demonstrated an approximate 15% improvement in network performance with intelligent 5G MAC scheduler and near-real-time on the RIC platform. NetAnticipate5G, along with Intel’s Analytics Zoo, running on Intel® Xeon® scalable processors, not only improved AI accuracy by 55% but also helped in reducing AI inference time to less than 1 msec (~ 0.56 msec). This combination of AI enablers provided over 41% improvement[1] in inference time , which is one of the critical requirement for such ultra-low latency and heighly reliable machine learning (ML) based solutions.

Capgemini Engineering is part of these alliances and is developing RIC platform enablers. We off er our awardwinning NetAnticipate5G framework that introduces AI as part of the RIC to support autonomous intelligent operations. Also, RATIO is a Capgemini Engineering RIC platform based on O-RAN Alliance specifi cations for developing a disaggregated intelligent RAN controller.

By introducing ML-infused Link adaptation, we observed gains in the spectral effi ciency of the order of 15% that resulted in cell throughput gain of 11.76%. Also, Project Marconi demonstrated that ML could be deployed for intelligent decision-making as part of RAN Layer 2 timecritical functions, on standard Intel® Xeon® processors, without introducing additional hardware requirement for CSPs.These are promising results for CSPs that could help them enhance the user service experience and reduce churn while improving radio resource utilization.

This instills confi dence in the Project Marconi team to continue with the research and introduce ML to infuse intelligence and enhance the performance of multi-user MIMO and massive MIMO scheduling functions.

Our plan now is to showcase our initiatives through the O-RAN Alliance and work with the community to add these concepts and their implementation approach to the O-RAN specifi cations Work Group items.

Please fill up the form to download the whitepaper