Adaptive Quantification of Noise Variance Using Subspace Technique: Fid-Bau Portal

Adaptive Quantification of Noise Variance Using Subspace Technique

Li, Z. / Chang, C. C.

Abstract

Filters are commonly used in the postprocessing of noise-contaminated measurement signals for achieving noise removal or data fusion. Quantifying variance of noise embedded in the signal is an important task for the calibration and operation of filters. In this study, an adaptive subspace technique is proposed to identify and track noise variances with unknown and potentially time-varying characteristics from noise-contaminated signals. The technique is developed on the premise that the covariance of the clean signals is rank-deficient, i.e., the number of excited vibration modes is less than that of measured channels. After separating the vector space of noisy signals into a signal subspace and a noise subspace, noise variances are estimated using the energy of the noise subspace. The projection approximation subspace tracking technique is then employed to adaptively update the signal subspace and track the changes of noise variances from the measurement data. A numerical example of a 10-story building model and a field test on a pedestrian bridge are performed to validate the proposed technique. Results show that the proposed technique can identify and track time-sensitive noise variances accurately and efficiently from noisy measurement data. It is also illustrated that the proposed technique can be integrated into the Kalman filter for fusing acceleration and displacement data and provides a better displacement estimate when the data contain time-varying noises.