Traditional noise-filtering techniques are known to significantly alter features of chaotic data. In this paper, we present a noncausal topology-based filtering method that is effective in removing additive, uncorrelated noise from time-series data. Signal-to-noise ratios and Lyapunov exponent estimates are dramatically improved following the removal of the identified noisy points.
Full paper in gzipped postscript (2MB) and pdf (4MB).