Deep Learning Framework for Spectral Efficient Intelligent Hybrid Beamforming

Abstract

Next-generation wireless networks’ attractive use cases call for more extensive coverage and highly dependable connectivity. A promising candidate that considerably helps to fulfil these requirements is beamforming. In massive Multiple-Input-Multiple-Output (MIMO) systems, the conventional digital beamforming method results in significant costs and hardware complexity. By using fewer RF chains than the conventional digital beamforming method, hybrid beamforming lowers the hardware needed. However, due to the restrictions on hardware consumption, it is difficult to arrive at the open optimal solution for joint optimization problems. We suggest a hybrid beamformer that learns to maximize spectral efficiency using deep learning as its foundation. To achieve the optimal beamforming weights, the channel state information (CSI) is supplied into the deep learning model. Both perfect and imperfect CSI are used to validate the proposed hybrid beamforming scheme. Simulation results reveal that the proposed method outperforms the current statistical approaches while lowering cost and hardware complexity. It is also more robust to poor CSI.