A new nonparametric method for linear system phase recovery from bispectrum

For input signals with nonzero third order moment, the phase of a linear system transfer function can be identified from the bispectrum of its output signal. Instead of using partial bispectrum phase information which is more sensitive to noises and measurement errors, a new nonparametric method is proposed to utilize all the available bispectrum phase information. The proposed new method is computationally simple. It does not require finding the pseudo-inverse of a gigantic matrix as some existing methods do. Its effectiveness is demonstrated through computer simulations