Rule of two: methods to detect and localize non-stationary noise in sweep measurements

Exponential swept-sines are currently one of the most popular excitation signals in acoustic measurements, being praised for their favorable signal-to-noise ratio and ease in rejecting harmonic distortion artifacts. However, they are notoriously susceptible to non-stationary noise, such as randomly occurring transients or noise bursts. Such disturbances cannot be mitigated in the deconvolution process, leading to low quality of obtained impulse responses and errors in acoustic parameter estimation. In this paper, we present our work on non-stationary noise detection and localization in measured sweeps, methods we named Rule of Two (Ro2) and Short-term Rule of Two, respectively. In both techniques, we employ correlation to evaluate the similarity of measured signals, while considering expected contamination: background noise and time variance. Comparing correlation coefficients of a series of measurements allows the selection of a clean pair, devoid of non-stationary noise in Ro2. If a sweep contains a disturbance, we can pinpoint the contaminated region by computing its windowed cross-correlation in Short-term Ro2. Validation on a large set of sweep measurements proves the reliability and robustness of both methods. The Ro2 and Short-term Ro2 methods enable robust and efficient acoustic impulse-response measurements in noisy places.