Median-based noise floor tracker (MNFT): robust estimation of noise floor drifts in interferometric data

The sensitivity of a search algorithm is determined by the noise floor, hence it is important to track drifts in the noise floor. For example, in the case of externally triggered search, one needs to look at long stretches in time, so it is imperative to check if the noise floor has drifted during that time. A method (called the median-based noise floor tracker or MNFT) has been developed to track the slow drifts in the noise floor. The method is robust against transients and any artefact mimicking non-stationarity that a data conditioning procedure may introduce (e.g., high-order notch filters may spread clustered transients). MNFT, which uses running median as the test statistic, smoothes the data to eliminate transients in contrast to methods such as running mean, where transient interference cannot be removed. Thresholds are set from simulations and any excursion of the test statistic beyond the threshold serves as an indicator of non-stationarity. The method is illustrated with the use of E7 and S1 data from an uncalibrated channel of the LIGO detector. MNFT can track even small fluctuations in the variance.