Adaptive bispectral estimation using Thomson's multitaper approach

We have compared the bias and variance properties of multitaper based bispectral estimators, with focus on an adaptive approach. Numerical results for a white Gaussian process and a nonGaussian AR(2) process are presented We found that the bias and variance properties of the multitaper bispectral estimators are governed mainly by a quantity we call the total bispectral window and by the leakage properties of the estimator. We conclude that the adaptive multitaper bispectral estimator represents the best compromise among all nonparametric bispectral estimators.