Since the number of independent array data snapshots is limited by the availability of real-world data, we propose a parametric bootstrap for resampling. The proposed parametric bootstrap is based on a generative adversarial network (GAN) following the generative approach to machine learning. For the GAN model we chose the Wasserstein GAN with penalized norm of gradient of the critic with respect to its input (wGAN\_gp). The approach is demonstrated with synthetic and real-world ocean acoustic array data.

generative adversarial network; GAN; sample covariance matrix; DOA estimation; Hellinger distance