Paper detail

Deep Learning based Joint Precoder Design and Antenna Selection for Partially Connected Hybrid Massive MIMO Systems

Efficient resource allocation with hybrid precoder design is essential for massive MIMO systems operating in millimeter wave (mmW) domain. Owing to a higher energy efficiency and a lower complexity of a partially connected hybrid architecture, in this letter, we propose a joint deep convolutional neural network (CNN) based scheme for precoder design and antenna selection of a partially connected massive MIMO hybrid system. Precoder design and antenna selection is formulated as a regression and classification problem, respectively, for CNN. The channel data is fed to the first CNN network which outputs a subset of selected antennas having the optimal spectral efficiency. This subset is again fed to the second CNN to obtain the block diagonal precoder for a partially connected architecture. Simulation results verifies the superiority of CNN based approach over conventional iterative and alternating minimization (alt-min) algorithms. Moreover, the proposed scheme is computationally efficient and is not very sensitive to channel irregularities.