Deep Learning in the 6G Air Interface

Back in May, Samsung Electronics hosted their first 'Samsung 6G Forum' (S6GF) that I blogged about here. The talk by Prof. Jeffrey Andrews, The University of Texas at Austin, deserves its own separate post. The topic of his talk was 'Deep Learning in the 6G Air Interface'. Video at the bottom of this post.

Quoting from Mobile World Live

In a presentation, Andrews noted emerging 5G applications including autonomous vehicles and robots require situational awareness going beyond what they can sense alone.

“Although driverless cars are built to be autonomous, they don’t really work right unless they can see things and know about things outside of their own field of vision. Otherwise, they’ll have to drive too slowly, too conservatively.”

Andrews said the push behind 6G is driven by increasingly data-hungry use cases, citing projections mobile network traffic will increase by up to 50-times by 2030.

He highlighted cost control as another major research theme, noting denser base station deployments would likely require an unprecedented amount of sharing and cooperation, and reuse of infrastructure across different operators.

Andrews cautioned it could be hard to squeeze out better performance in many areas of 6G than current 5G networks, noting the latter technology’s physical layer was developed over decades and the industry was already at an advanced stage on theoretical and implementation pathways.

Andrews believes machine learning will boost site-specific learning and design, with much of his research focused on improving beam management.

He noted deep learning is a powerful tool for wireless development, but it’s not a panacea, adding learning when and how to use it is a major research challenge for the next decade.

Worth highlighting his last point that good datasets and publicly available simulators are very important for ML-wireless research, this is a big challenge on which industry and academia should co-operate.

Also check out the 6G@UT's research page. As you can see from the slide above, they are focussing on the following four key areas:

  1. Deeply Embedded Machine Learning
  2. New Spectrum and Topologies
  3. Pervasive Sensing
  4. Open Networks

The video of his talk as follows:

Related Posts