Zero Forcing Beamforming With Sidelobe Suppression Using Neural Networks

The use of deep learning in the field of wireless communications has already shown great potential. In this work we present a deep feedforward and a deep recurrent neural network trained as null steering beamformers that target high sidelobes in order to establish low sidelobe level for any desired incoming signal. Using of a zero-forcing algorithm, we apply a sidelobe-damping algorithm where iteratively, a constant number of sidelobes is nullified until a desired sidelobe level is reached. In this way, we can create a dataset which we later use to train our NN models. Using Bayesian optimization, we perform hyperparameter tuning to configure structure and training related parameters for the NNs under examination. The NN models are later fine-tuned using a small dataset containing more demanding cases.