Adding intelligence to the radio access network is the focal point of AI and ML adoption by telecom operators and their vendors. And given the increases in traffic and complexity brought about by the advent of 5G, alongside the move from proprietary to virtualized/cloud-native network functions, this makes perfect sense.
According to Dell’Oro Group Vice President Stefan Pongratz, “The increased complexity with the various 5G technologies in combination with the shift towards Open RAN will potentially introduce new challenges to CSP operational teams tasked with managing end-to-end performance. Artificial intelligence will play an increasingly important role managing this complexity deliver the quality of experience (QoE) that consumers and enterprises demand from mobile broadband applications and latency-sensitive services.”
Ericsson is working with its customers to implement AI and other automation capabilities to monitor network performance and take predictive action that, if not addressed, would result in a dip in QoE. In an interview with Light Reading, Ericsson’s Head of Automation and AI Jonas Åkeson said RAN optimization and root cause analysis are among the AI-related use cases the vendor is putting into production.
“Customer experience is super important for all of us,” he said in the interview. “Now we are predicting way ahead of time. Two hours in advance we can see that the network is starting to lean towards degrading…We can actually take action before it has customer-impacting alarms coming in. We are now closed-loop so we can see things in advance that triggers our automation capability to start sending out commands into the network. We are improving throughput, latency, and the customer experience in the range of 25-30%.” For root cause analysis, he said it used to take around a week to examine eight performance indicators on 100 cells. Now, in addition to taking a more nuanced view of performance, the same number of cells can be studied in 5 minutes and the entire network in 15 minutes. “We are doing things now which is previously humanly impossible. This is really a game changer we believe.”
In early 2021, Nokia and China Mobile completed live trials of AI for RAN applications on the carrier’s 4G and 5G network. Specifically the pair tested an AI-based real time UE traffic bandwidth forecast in Shanghai, and automated network anomaly detection in Taiyuan. A RAN Intelligenct Controller (RIC) was deployed in edge cloud infrastructure.
In Shanghai, the trial confirmed that AI-based real-time user equipment (UE) traffic prediction accuracy exceeded 90% in a live 5G network test. With the real-time RAN data exposure capability, Nokia’s 5G AirScale base station was able to send UE radio quality information to the RIC in real-time, which is critical for the accuracy of the predictions.
“RIC plays a key role in enabling AI/ML capability in the RAN,” Huang Yuhong, deputy director of China Mobile Research Institute, said. “Nokia and China Mobile’s trials are very meaningful for RIC commercialization. China Mobile has put effort into the AI-assisting RAN network technology. We are pleased to complete these trials using AI to forecast UE transmission bandwidth and detect anomalies on China Mobile’s live network…The field trial proved the availability of RIC enabling network enhancements through customized real-time BTS data analysis and control.”
RIC technology has been standardized through the O-RAN Alliance. There are two flavors: the near real-time RIC and non-real-time RIC. The near real-time RIC hosts microservice-based applications referred to as xApps for managing and optimizing the distributed RAN components–eNodeB, gNodeB, central unit and distributed unit. Data from RAN elements including macro sites, massive MIMO arrays and small cells, are passed through the near real-time RIC where xApps analyze and optimize for user experience and resource utilization.
The non-real-time RIC handles things like configuration, device, fault, performance and lifecycle management. This lets new radio units be self-configured rather than manually configured which is important given the density necessary as 5G deployments continue. Operators can tap non-real-time RICs for policy-based guidance and for AI and ML training.
In the context of multi-vendor Open RAN configurations, the ability for ongoing, autonomous RAN management and optimization enables 5G to be adapted for a wide range of deployment models and the services that flow from that. According to Parallel Wireless’s Eugina Jordan, “The RIC platform provides a set of functions via xApps and using pre-defined interfaces that allow for increased optimizations…which leads to faster and more flexible service deployments and programmability within the RAN. It also helps strengthen a multi-vendor open ecosystem of interoperable components for a disaggregated and truly open RAN.”
VMware’s VP of Telco Strategy Sachin Katti said the goal is to combine edge and RAN so “eventually applications can leverage the cloud all the way to the edge to run things you cannot do today. Apart from having a cloud platform that you can run these cloud applications on at the edge, the other thing we’re focusing on is these applications don’t just sit alongside the RAN; they can interact with the RAN.”
The post AI and ML for RAN optimization, automation appeared first on RCR Wireless News.