AI/ML in 5G RAN and Core

This IDC Market Perspective focuses on the use of AI and ML in the development and operations of the 5G RAN including O-RAN and the 5G core. The cellular industry began working with AI/ML four to five years ago, starting with 4G long-term evolution (LTE) and has developed a handful of algorithms that can be used to address congestion, beam forming, and radio optimization. As operators around the globe start to deploy 5G, the RAN and core network equipment providers will continue to evolve their AI/ML capabilities for 5G. Owing to the sheer complexity of over 2,000 network parameters in 5G, which need to be managed and optimized, combined with the introduction of network slicing, multi-access edge computing (MEC), and ultra-reliable low-latency communications (URLLC), AI/ML might evolve into the most practical way to solve complex optimization problems. AI/ML can be built on top of the RAN or it can be embedded in the RAN, with the latter needing additional compute capacity in the baseband processor. AI/ML in the 5G core can be embedded into the VNF servers as needed, and this would apply to both the on-premises and cloud-based implementations of the 5G core."Rel-18 will serve as a platform to expand AI/ML to the end-to-end system across RAN, device, and air interface and through data analytics expand capabilities of the core with AI/ML. RAN use cases that are expected to be enhanced and modeled include energy efficiency, self-optimizing network (SON), channel state information (CSI) feedback, load balancing, and massive MIMO beam management," says Bill Rojas, adjunct research director, IDC's Asia/Pacific Communications Group.