Deep Learning with Interference Training for Adaptive Radar Beamforming

Phased arrays electronically shape and steer elec-tromagnetic beams for transmission or reception of signals. The basic operation of beamforming can be expressed as a tensor operation, thereby laying the foundation of using deep neural networks for adaptive beamforming in radar systems. In this paper, we formulate adaptive receiver beamforming under interference effects as tensor operations and train a deep neural network (as a regression model) along with interference training to boost beamforming capabilities and mitigate interference effects in radar systems. The performance results indicate that our formulation operates with small memory footprint and low latency in inference time, while effectively suppressing both noise and interference (such as barrage jamming) effects.