Researchers propose a deep unfolding approach for joint communications and sensing hybrid beamforming in 6G networks to overcome the high computational cost of real-time operation in high-frequency bands. This method converts classical optimization iterations into trainable neural network layers, embedding domain knowledge to allow the analogue and digital precoder components to update at their respective optimal timescales. The resulting nested architecture achieves higher communication rates and more accurate sensing patterns while converging significantly faster than conventional techniques. By maintaining the mathematical structure of traditional algorithms while learning improved update rules, the system offers a practical solution for resource-constrained, large-scale 6G deployments.
Source: https://www.6gflagship.com/news/beamforming-for-networks-that-sense-and-communicate-at-once/
Keywords: deep unfolding, hybrid beamformer, model-based machine learning, joint communications and sensing, alternating optimization