Polyspectral analysis of (almost) cyclostationary signals: LPTV system identification and related applications

Polyspectral estimators are proposed for (almost) cyclostationary signals and are shown to be consistent and asymptotically normal. These estimators are used for identification of linear (almost) periodically time-varying systems. Both nonparametric and parametric approaches are described for input-output and output only identification. Statistical analysis of nonstationary signals with missing observations is treated and tests are developed for checking the presence of cycle frequencies. Frequency estimation and detection of coupling are addressed in the cyclic domain without resorting to phase randomization. All methods are proven to be insensitive to stationary noise and use consistent single record estimators