Deep learning driven beam selection for orthogonal beamforming with limited feedback

This letter studies deep learning methods for beam selection in multiuser beamforming with limited feedback. We construct a set of orthogonal random beams and allocate the beams to users to maximize the sum rate, based on limited feedback regarding the channel power on the orthogonal beams. We formulate the beam allocation problem as a classification or a regression task using a deep neural network (DNN). The results demonstrate that the DNN-based methods achieve higher sum rates than a conventional limited feedback solution in the low signal-to-noise ratio regime under Rician fading, thanks to their robustness to noisy limited feedback.