How close are we to fully-automated Open RAN networks?

Open RAN has been the subject of earnest development for a number of years and has seen large-scale commercial deployment by three major global operators building new networks with significant commitments from many large brownfield CSPs. Reducing TCO through vendor competition, component commoditization and cloud-scale are major investment drivers but there’s also the promise of new types of innovation derived from the RAN Intelligent Controller. As a cloud-based RIC ingest network telemetry, it can provide near-real-time network visibility and optimization capabilities that further the move toward network automation. But how close are we to fully-automated Open RAN networks? 

In a panel discussion at the Open RAN European Forum, LightCounting Principal Analyst Stelyana Baleva talked through the current state of network automation, the future outlook and the role of Open RAN with industry experts. While there was agreement that network complexity is rapidly increasing which raises additional challenges, the opportunity is there and that’s a good thing as operator’s face lack of differentiation and stagnant service revenues that can be combated by end-to-end automation. 

Open RAN extends programmability

Henri Helanterä, a business manager with Elisa Polystar, an offshoot of Finish operator Elisa specialized in network automation solutions for CSPs, explained: “I think Open RAN brings a lot of capabilities to programmatically manage the radio access network at the level that has not been possible in the past, but then at the same time, this disaggregated architecture, so both horizontal disaggregation and vertical disaggregation, it also adds some complexity that needs to be addressed. So it brings some solutions, but it also brings some challenges for zero touch operation of the network.” 

CommScope’s Colin Bryce, senior director of product line, tracked the movement from self-configuration, self-healing and self-optimization into the world of software-based service management and orchestration in virtualized and disaggregated networks. “I think it really is about end-to-end orchestration of the network and then controlling the physical layers and the software implementations to deliver those services and manage them automatically,” he said. 

As for the role of Open RAN, it’s not strictly necessary, Bryce said but, “I think openness and particularly defining data structures and defining protocols makes it much, much easier to allow the concept of network management to be implemented end-to-end across the network. “

The role of the RIC in Open RAN network automation

Bryce called the RIC “one of the more important aspects of the Open RAN architecture” in terms of automation capabilities. He called out development of xApps and rApps that could run on the RIC and focus on specialized applications as filling a gap in current telecom domain expertise. “Let’s face it: most of us in the mobile industry today are telecoms experts. We’re not AI experts. We’re not machine learning experts. But if we can deliver the APIs and we can deliver the structures of data to guys who really are experts in those fields, they will be able to drive those self-learning algorithms into our industry.” 

xApps and rApps hosted on the RIC, Helanterä said, cover control of handover, dual connectivity, carrier aggregation, and potentially beamforming, will need “a standardized platform or framework created for running these applications…These applications could come from different providers, and therefore, those application providers could focus on the application logic without having to recreate all the data management and security infrastructure.” 

Exposing Open RAN data for further RIC development

In terms of the RIC in the broader context of automation and its maturation, Helanterä gave a bit of a reality check. “There are of course a lot of loaded expectations that machine learning will bring some kind of a magic into the way that we manage these networks and the way we automate these networks by creating adaptable and generalized logic that can span across different kinds of network deployments without a need to tweak the logic between different instances.” 

To get to that point, it’s important, he said, to think practically and focus on exposing data for use in machine learning model development which further requires modern data platforms and good data management. “I think these kind of architecture initiatives that we see with the O-RAN Alliance, for example, that is certainly taking us to the right direction there.” 

Bryce noted the delta between the technological ability to automate a network function and operators’ comfort levels with putting that into real-world practice. The first step toward building confidence will be delivering on specified KPIs to demonstrate, “Hey, we can take the human out of the loop, as it were, and let these machine algorithms begin to manage the network. But, I think, psychologically, for a lot of network operators, I think there’s going to have to be that initial step of not just turning this on and hoping for the best and seeing that it improves, but actually comparing the performance with some idealistic plan and giving us the confidence that this really is going to bring the benefits that we see.”