Abstract
Removal of narrowband noise features (also called lines) of known instrumental origin from a time series is important for improving the performance of algorithms, such as those for the detection of transients. We present a new method for removing lines which (i) is not based on any model for the features to be removed, (ii) is designed so as not to affect transients substantially and (iii) works in the time domain. Property (i) allows lines to be removed irrespective of their physical origin, (ii) ensures that transients remain detectable in the residual after line removal and (iii) means that, unlike Fourier domain methods, line power is not redistributed in the entire frequency band.