A basic study of noise reduction on the analysis of burst gravitational waves by direct and parallel denoising autoencoder

Reconstruction of a burst gravitational waveform by denoising observed noisy data is one of the essential issues of gravitational wave astronomy. Conventional denoising methods require the knowledge of the frequency bands of the noise which is contained in observed data, but it is difficult to understand the statistical properties of the noise of observed data because it is known that the noise has non-Gaussian and non-stationary properties. In this paper, we propose direct and parallel denoising autoencoder for high-quality denoising without such kind of knowledge, and demonstrate our algorithm to reconstruct one of the typical models of burst gravitational waveforms from the noisy data.