Nokia Bell Labs and TU Dresden

Employing deep learning to address network slicing issues

DAP TU Dresden ResrcID23781_GettyImages-168953931-v2.jpg
DAP TU Dresden ResrcID23781_GettyImages-168953931-v2.jpg

Nokia Bell Labs, TU Dresden employ deep learning to address network slicing issues

Why build multiple networks when the existing one can be shared? That’s the basic concept behind 5G network slicing. 5G slicing allows multiple parties to enjoy the benefits of having their own private network, while sharing the same physical resources.

But slicing has two major challenges. The first is ensuring that the network slice delivers the performance guaranteed in customers’ Service Level Agreements (SLAs). Traditionally, this has been achieved by giving each customer its own isolated slice of spectrum. However, the practice of isolating slices often results in significant amounts of bandwidth being underutilized. The second challenge, therefore, is making the most efficient use of the network’s radio spectrum, which requires multiplexing, or allowing multiple parties to use the same common resources.

To help solve this conundrum, Nokia Bell Labs and Technische Universität (TU) Dresden have developed a deep-learning approach for dynamically controlling and adjusting the network’s Radio Resource Management (RRM) mechanisms. The new methods enable the RRM to multiplex spectrum across slices while ensuring that each customer’s SLA is upheld.

“Because it isn’t dependent on a particular model, this deep-learning tool can automatically adjust how spectrum is allocated as KPIs and bandwidth demands change,” said Dr. Gerhard Fettweis, professor at TU Dresden. “Once it knows the rules of the game, it can determine the most strategic way to play.”

In a series of lab simulations, researchers introduced a wide range of load anomalies to the network to test its response. The researchers have found that, by using deep learning methods instead of heuristic-based algorithms, the network could tolerate 43% more load anomaly caused by the overloading slice without affecting the KPIs of other slices.

The team also plans to investigate how the user equipment in overloaded cells might be handed over to underloaded cells to better fulfill SLA requirements. While similar load-balancing methods proved only moderately effective in the past, the team anticipates that the use of multi-panel user equipment will provide much greater benefits since such panels introduce interference-isolation between neighboring cells. This approach might be particularly appealing for 5G high-carrier frequency systems where multi-panel user equipment will have high penetration.

Nokia Bell Labs and TU Dresden have collaborated on several research projects over the past few years. These projects have focused on various LTE and 5G radio system communication aspects in the field of radio resource, interference and mobility management.