Time-series analysis (Q3988348)

This paper addresses the problem of identifying and characterizing strange attractors arising from non-linear evolution equations by using time series of measured values. In particular, the authors consider time series of voltage measurements taken from an electronic non-linear oscillator with analog data converted to a 12 bit digital form at a sampling frequency of 10 kHz. Quantization noise obscures the existence of the strange attractor, and classical signal processing techniques are not helpful at separating noise from the complex fractal signal. The authors devise a strategy for isolating and extracting geometrical information from the time series. They use the notion of a delay register to convert the time series into a static set embedded in \(\mathbb{R}^ n\) for some fixed \(n\). One then assumes that this set has been generated by sampling a measure that is invariant under the time evolution of the system. The object then is to characterize the geometry of the invariant measure.