Two-dimensional Direction Estimation of Generative Adversarial Beamforming

To solve the problem of large main lobe width of conventional beamforming methods, a two-dimensional deconvolution beamforming method based on machine learning is proposed. The algorithm uses the conditional generative adversarial network to learn the mapping relationship between beam power spectrum and point source distribution. In the process of algorithm training, the two-dimensional scanning results of conventional beamforming are input. The main lobe of the network output is narrower. Since this method inherits the advantages of the conventional beamforming, it is more robust than the adaptive beamforming method in the case of steering vector errors.