Removing non-stationary, non-harmonic external interference from gravitational wave interferometer data

TL;DR

This paper presents an adaptive method to identify and remove non-stationary, non-harmonic electrical interference lines from gravitational wave detector data, improving signal clarity and enabling detection of previously masked signals.

Contribution

The authors develop an empirical model and an adaptive algorithm to effectively cancel complex electrical interference in gravitational wave data, even without direct reference signals.

Findings

Interference lines are successfully removed from detector data.

Recovered signals retain at least 75% of their original amplitude.

The method enhances the spectral cleanliness of gravitational wave data.

Abstract

We describe a procedure to identify and remove a class of non-stationary and non-harmonic interference lines from gravitational wave interferometer data. These lines appear to be associated with the external electricity main supply, but their amplitudes are non-stationary and they do not appear at harmonics of the fundamental supply frequency. We find an empirical model able to represent coherently all the non-harmonic lines we have found in the power spectrum, in terms of an assumed reference signal of the primary supply input signal. If this signal is not available then it can be reconstructed from the same data by making use of the coherent line removal algorithm that we have described elsewhere. All these lines are broadened by frequency changes of the supply signal, and they corrupt significant frequency ranges of the power spectrum. The physical process that generates this interference is so far unknown, but it is highly non-linear and non-stationary. Using our model, we cancel the interference in the time domain by an adaptive procedure that should work regardless of the source of the primary interference. We have applied the method to laser interferometer data from the Glasgow prototype detector, where all the features we describe in this paper were observed. The algorithm has been tuned in such a way that the entire series of wide lines corresponding to the electrical interference are removed, leaving the spectrum clean enough to detect signals previously masked by them. Single-line signals buried in the interference can be recovered with at least 75 % of their original signal amplitude.