Transfer function phase estimation using a low variance higher order spectrum

A method is presented for identifying a linear system from observations of the output only. The authors introduce a new one-dimensional spectrum, P( omega ), and use it to obtain the underlying linear system. The proposed method incorporates builtin smoothing mechanisms to reduce the variance of the phase. The proposed approach requires fewer computations that the Lii-Rosenblatt approach (1982), and it also provides a new way to look at higher-order spectra.<<ETX>>