Median-based cyclic polyspectrum estimation

In this paper, median-based estimation methods for the cyclic polyspectrum are proposed. The algorithms do not require a priori knowledge of the β submanifolds, that is, they do not require the knowledge of all of the lower order cycle frequencies of the time-series available for the estimation of the cyclic polyspectrum. Therefore, such methods are particularly useful when the cyclostationarity of the signals under consideration is not completely known. The proposed estimators converge to the theoretical values of the cyclic polyspectrum when the collect time approaches infinity and the spectral resolution becomes infinitesimal. Furthermore, their accuracy is very nearly the same as that of the usual time- and frequency-smoothed cyclic periodogram methods that use a priori knowledge of lower-order cycle frequencies to avoid the β submanifolds