Filtering of randomly occurring signals by kurtosis in the frequency domain

Abstract

m e paper describes a nonlinear operator, based on Frequency Domain Kurtosis (FDK), that is able to distinguish between transients (impulses and unsteady harmonic components) and stationary sinusoidal signals in background Gaussian noise. For problems involving signal detection in burst noise, filtering by an FDK operator may improve detection results by reducing noise variance and amplifiing narrow-band transient signals.To evaluate the FDK operator's eflciency, direrent processes were considered: impulsive noise and narrowband inteferences in background Gaussian noise and real electromagnetic non-Gaussian noise. The filter robustness is discussed.

The spectral kurtosis: a useful tool for characterising non-stationary signals

“…[7], for the specific case where Y ðtÞ is a linear combination of pure tones. More generally, it is worth insisting on the important potential of Property 10 in detection problems.…”

“…Since then, the SK had been seldom brought into play [5], until Pagnan and Ottonello proposed a modified definition based on the normalised fourth-order moment of the magnitude of the short-time Fourier transform [6,7]. This led to considerably simplified properties.…”

Spectral kurtosis for fault detection, diagnosis and prognostics of rotating machines: A review with applications

“…In 1983, frequency domain kurtosis (FDK) was first developed as the kurtosis of its frequency components in the frequency domain by Dwyer [22], and then it was used as a complement to the power spectral density to detect "randomly occurring signals" in [23,24]. In 1994, Pagnan based on the normalized fourth-order moment of the magnitude of short-time Fourier transform (STFT) [25,26]. They also showed that SK could be used as a filter to recover random signals even when they are severely corrupted by additive stationary noise.…”

“…An estimator of the SK based on the STFT was originally suggested in [22][23][24][25][26][27], while its explicit deduction from a timefrequency approach was given in [29,42,43]. For a process YðtÞ with an analysis window wðnÞ of length N w and a given temporal stepsize P, the STFT is written as…”

Meteor detection and non-periodic bursty interference removal for Arecibo data

“…We use the power profile of each IPP signal to do the interference detection because it is always real and nonnegative. We calculate the central moments (similar techniques apply in frequency domain can be found in Pagnan et al, 1994) and apply a nonlinear filter to get the power reduction percentage (PRP) of the power profile (the power profile and PRP are defined in Section 2). Using these parameters we identify the presence of the interference and then blank the corresponding signal samples.…”