On Derivation of Bispectra for Stationary Stochastic Process

Based on an artificially modeled realization of a random process $\xi_k$, $k=\pm 1,\dots$, with a prescribed bispectral density $g(\lambda_1,\lambda_2)$, we construct and compare various estimates of the bispectral density. We show that using high-order weight functions in constructing an estimate for the bispectral density considerably reduces the estimation error.