DeepTx: Deep Learning Beamforming with Channel Prediction

Machine learning algorithms have become widely popular in wireless communications. Previously, we have proposed to use a deep fully convolutional neural network (CNN) to receiver processing and showed it to provide promising gains in receiver performance. In this study, we focus on machine learning algorithms for transmitter. In particular, we consider beamforming and propose a CNN which, for given uplink channel estimate as input, outputs downlink channel information for zero-forcing beamformer. The CNN is trained in supervised manner considering both uplink and downlink simulated transmissions with a loss function that is based on UE receiver performance. The main task of the neural network is predict the channel evolution between uplink and downlink slots, but it can also learn to handle inefficiencies and errors in the whole chain, including the beamforming itself. Numerical experiments are used to demonstrate greatly improved beamforming performance.