Statistical method of sky removal for submillimeter ultra-wideband spectrometers
Removing emission of Earth's atmosphere (sky emission) from observed data is essential to obtain sky-corrected astronomical signals for ground-based spectroscopy in the millimeter and submillimeter wavelengths. Since it fluctuates as 1/f-type, one would need to measure time-series spectra and distinguish sky emission from astronomical signals by any means. The conventional method is to switch the sky position and obtain two spectra, only one of which signals are entered in, by turns in order to cancel sky emission. As the instantaneous bandwidth are reaching tens or hundreds of GHz in the next ultra-wideband spectrometers, however, it would fail to cancel sky emission over the entire frequency range because of non-linear frequency dependency of it. In this talk, we present a statistical method to achieve sky removal for such instruments. Our approach is to obtain time-series spectra at faster sampling rate (10-100 Hz) than switching (0.1-0.2 Hz) and instantaneously fit and remove non-linear spectral curve caused by ultra-wideband sky emission after the conventional sky removal. We achieve the fitting by a linear regression with sparse regularization, where non-linear basis is calculated by an atmospheric model. With the data of DESHIMA, an ultra-wideband spectrometer based on on-chip filterbank and microwave kinetic inductance detector (Endo et al. 2019a, 2019b), we demonstrate that the proposed method significantly mitigates the non-flatness of an estimated astronomical spectrum compared to the conventional one.
