SoftBank has implemented a machine learning based method for radio access network design from Ericsson.
The service groups cells in clusters and takes statistics from cell overlapping and potential to use carrier aggregation between cells into account, thus reducing operational expenditure and improving network performance. Compared to traditional network design methods, it cut the lead time by 40 percent, claims Ericsson.
The foundation for the method is a thorough analysis of the actual radio network environment, for example taking cell coverage overlap, signal strength and receive diversity into consideration. The high number of possible relations between cells as well as considerations for network evolution, calls for substantial computational power and state-of-the-art machine learning techniques.
This highly complex task was a tremendous challenge that Ericsson solved by implementing a cutting-edge design concept based on network graph machine learning algorithm (community detection) that Ericsson has now patented.
Despite these challenges, SoftBank was able to automate the process for radio access network design with Ericsson’s service. Big data analytics was applied to a cluster of 2000 radio cells and data was analyzed for the optimal configuration.
Ericsson is combining extensive radio networks competence with the latest machine learning advances to provide a key differentiator for customers on the road to automation.
To support Ericsson’s global strategy to bring machine intelligence into different streams, Ericsson Network Design and Optimization is running an Artificial Intelligence Accelerator Lab, hosted in Japan and Sweden, looking to develop these use cases.