Implementation of Deep Learning in Beamforming for 5G MIMO Systems

Abstract

The fifth generation 5G communication systems depend on beamforming and massive MIMO techniques to enhance the performance. In massive MIMO architecture hybrid beamforming is typically adopted to reduce the required number of RF chains. We present a technique based on deep learning to simplify the process of estimating beamforming weights. First, a fading communication channel model is developed, and the generated data is used to train convolution neural networks. The trained networks are used to predict beamforming weights based on estimated channel data. Results are presented of the implementation of deep learning in digital as well as hybrid beamforming. Presented results reveal the potential of deep learning in reducing the complexity of estimating beamforming weights. The results also present a comparison of the performance the communication system depending on deep learning and conventional beamforming techniques.