The Role of AI in RAN Automation Posted on : Sep 13 - 2022

5G represents a tipping point within the telecoms industry where networks become too complex for humans to operate cost-effectively without the use of automation tools and technologies. Complexity is driven in part by 5G itself, which uses a much broader set of frequency bands, can prioritize services based on latency, and supports huge increases in the number of network elements and end-user devices. But there is a plethora of other changes which further increase complexity.

These include the evolution from physical hardware to virtual and cloud native networks, end-to-end network slicing, the adoption of Open Radio Access Network (RAN) technologies and the addition of new enterprise business services. There are also multi-technology networks with some communications service providers (CSPs) running 2G, 3G, 4G/LTE and 5G networks in parallel, as well as multi-vendor networks with typically two to four different RAN vendors deployed in the network.

Artificial intelligence (AI) and machine learning (ML) are becoming commonplace in the telecoms industry and are often the only way to manage the complexity we see in today’s multi-vendor, multi-technology networks. This complexity gets even more apparent in the RAN that are one of the most challenging domains to tackle due to its sheer distributed nature, number of network nodes and its proximity to the end users, which makes it very not surprisingly, a major consumer of OPEX.

RAN evolution embeds automation

The telecoms industry automation is strongly linked to the ubiquitous usage of AI – but where it makes more sense depending on the use case. For instance, improving CAPEX/OPEX rationalization and performance require actions in the network at scale. The good news is that the newest network technologies – 5G and Open RAN– have been designed for widescale automation. In fact, the O-RAN Alliance is defining a new service management and orchestration (SMO) architecture specifically designed to enable better RAN automation.

The key to success then is that network providers automate the right things and aim to continually improve performance, which requires applied intelligence. When it comes to the evolution of RAN automation, we can see AI and ML technologies used predominantly in three specific areas.

SMO platform – the SMO platform itself is designed to incorporate AI technologies. At its core it has an embedded AI/ML execution environment. The platform is designed to connect to large external data sources as well as support northbound and southbound interfaces.

Lifecycle management – we see an urgent need for more usage of AI in the lifecycle management of virtual and cloud native network software components. A key aim of RAN automation is to replace the manual work of developing, installing, deploying, managing, optimizing and retiring RAN functions. Because AI and ML have proven to be efficient tools to develop RAN automation functionality, the use of AI and ML to drive the lifecycle management and CI/CD tools is obvious. There is an expectation that AI and ML will be extensively used in the training and retraining of deployment models from the use of a generic, global model to a much more autonomous embedded model with autonomous retraining. Data collection and management is one of the biggest challenges to scale the AI/ML software and tools in the CSPs. It is utterly relevant how data is managed in the algorithm’s lifecycle. View more