Fast Inference Beamforming Prediction using Machine Learning

This work introduces an efficient, novel framework using a machine learning (ML) model to estimate beamforming weights on a Raspberry pi microcontroller. Tracking fast-moving targets and dynamically adjusting the weights of each antenna element in a phased array to transfer maximum energy to the target is desired in many beamforming applications. This work exploits ML models' fast output predicting nature to calculate optimum weights. A KNeighborsRegressor (KNR) is embedded into the proposed beamforming framework, which can predict the appropriate beam pattern of a 3×1 fractal array antenna. The process carried out on Raspberry pi has successfully calculated beamforming weights for 100 random targets within 62.8 seconds.