Noisy input/output system identification using integrated polyspectrum

Two new parametric, frequency domain approaches are proposed for estimation of the parameters of linear errors-in-variables models, i.e., linear systems where measurements of both input and output of the system are noise contaminated. Existing approaches based upon second-order statistics will yield biased estimates if noise statistics are unknown. The proposed approaches exploit an integrated bispectrum and cross-bispectrum of the input-output data where the integrated bispectrum of a signal is defined as a cross-spectrum between the signal and its square. The parameter estimators are shown to be statistically consistent and numerically globally convergent. One of the approaches is linear and the other is nonlinear. Computer simulation results are presented.<>