A frequency measure robust to linear filtering.pdf

a r X i v : n l i n / 0 3 0 5 0 3 8 v 1 [ n l i n .C D ] 2 1 M a y 2 0 0 3 A frequency measure robust to linear filtering A G Rossberg Zentrum fu?r Datenanalyse und Modellbindung, Universita?t Freiburg, Eckerstr. 1, 79104 Freiburg, Germany E-mail: axel@rossberg.net Abstract. A definition of frequency (cycles per unit-time) based on an approximate reconstruction of the phase-space trajectory of an oscillator from a signal is introduced. It is shown to be invariant under linear filtering, and therefore inaccessible by spectral methods. The effect of filtering on frequency in cases where this definition does not perfectly apply is quantified. PACS numbers: 05.45.Tp, 06.30.Ft, 05.45.Xt Submitted to: Nonlinearity , 22 January 2003 1. Introduction In the experimental exploration of complex systems, such es those encountered in life-science, geology, or astronomy, it is not unusual that the experimenter discovers oscillations of unknown origin in a measured time-series. The experimenter would then usually try to characterize these oscillations in a form that admits an identification of their source – the oscillator. The conventional first choice is a characterization of the oscillations by their “frequency”. For an ideal, periodically oscillating signal x(t), the smallest number T 0 such that x(t) = x(t + T ) for all t is the period of the signal, and its (angular) frequency is defined by ω = 2π/T . There is a good reason for choosing this particular characterization. All other properties of an ideal, periodic signal, i.e. its waveform and amplitude, are subject to distortions along the signal pathway from the oscillator to the detector. In fact, linear filtering along the signal pathway would generally be sufficient to modify the waveform and the amplitude in an arbitrary way. And the precise properties of the signal pathway are unknown in the setting considered here. The frequency information is the only sure fact. For ideal, periodic oscillations, these observa