Cyclic Spectral Analysis from the Averaged Cyclic Periodogram

The theory of cyclostationarity has emerged as a new approach to characterizing a certain type of nonstationary signals. Many aspects of the spectral analysis of cyclostationary signals have been investigated but were essentially based on the use of the smoothed cyclic periodogram. This paper proposes a cyclic spectral estimator based on the averaged cyclic periodogram which benefits from better implementation properties. It shows that an unexpected but important condition for this estimator to be valid is to set enough overlap between adjacent segments in order to prevent cyclic leakage. It proves that setting the percentage of overlap to 75% with a hanning window, or 50% with a half-sine window fixes the problem. It also shows that in certain situations the cyclic leakage associated with the averaged cyclic periodogram can be made exactly zero, in contrast with the smoot.