Adaptive filtering techniques for interferometric data preparation : removal of long - term sinusoidal signals and oscillatory transients

Received (received date) Revised (revised date) We propose an adaptive denoising scheme for poorly modeled non-Gaussian features in the gravitational wave interferometric data. Preliminary tests on real data show encouraging results. 1. Motivation Several large-scale interferometric gravitational wave detectors will come on-line soon, such as LIGO in the U.S., the French/Italian Virgo project, GEO600 the German/British interferometer and TAMA in Japan. Gravitational wave detectors produce an enormous volume of output. Data analysis techniques will have to be developed to optimally extract the weak signature of a gravitational wave from these data. Many of the techniques developed so far are based on matched filtering and assume stationary Gaussian noise. However, the real data stream from the detectors is not expected to satisfy the stationary and Gaussian assumptions. This disparity between standard Gaussian assumptions and real data characteristics poses a major problem to the direct application of matched filtering techniques in particular for burst sources such as black hole binary quasinormal ringings 1 or inspiral waveforms 2. In fact, the data from the Caltech 40 meter prototype interferometer has the expected broadband noise spectrum, but superposed on this are several other noise features: such as long-term sinusoidal disturbances coming from suspensions and electric main harmonics and also ringdown transients occurring occasionally, typically due to servo-controls instabilities or mechanical relaxation in suspension system etc. While no precise model can be given for this noise until the detector is completed and fully tested, matched filtering techniques cannot be used to locate/remove these noisy signals. We propose a denoising method based on adaptive linear prediction techniques 1 2 Adaptive filtering techniques for interferometric data. .. which does not require any precise a priori information about the noise characteristics. Although our method does not pretend to optimality, we believe that its simplicity makes it useful for data preparation and for the understanding of the first data. In the following, we present the structure of the proposed algorithm and some results obtained with the data from the Caltech 40 meter prototype interferometer 3. For a more detailed presentation, we refer the reader to 4. 2. Adaptive linear prediction The idea is to predict the current signal sample x k with a collection of past samples X k = (x k−d−n , n = 0, 1,. .. , N − 1) t , the delay d ≥ 1 being fixed arbitrarily. This is possible, only if the target …