Reliable and Efficient Beam Selection Using Conformal Prediction in 6G Systems

Massive MIMO (mMIMO) and Distributed MIMO (D-MIMO) systems are technologies that increase the capacity of next generation cellular systems. They have large number of beams or antennas from which the UE needs to choose for service. We focus on the D-MIMO system and propose techniques for choosing the optimal beams. Brute-force beam selection is not feasible in D-MIMO systems due to large number of necessary measurements. Machine learning (ML) models have been used in previous studies to reduce number of measurements and feedback. We propose Conformal Predication based approaches that enhance the models by making them more efficient and reliable. We also enhance the models with measurements taken at near-by locations to improve the predictions and validate our approach with simulation results.