Beamforming Designs for Hybrid Relaying in mmWave Systems Based on Deep Unfolding

In this paper, an efficient hybrid beamforming scheme is proposed for the relay node in mmWave relay systems based on deep unfolding. The optimal digital beamformer at the relay node is decomposed into an analog-digital- analog (A-D-A) hybrid structure, which can be formulated as a matrix factorization problem. The analog beamforming matrices and digital beamforming matrix in the A-D-A hybrid structure are efficiently solved by a deep unfolding neural network, which is unfolded based on the projected gradient descent algorithm. Numerical results verify that performance of the proposed scheme is close to the fully-digital solution, which prevails traditional model-based algorithms with a fast runtime.