Acousto-optic estimation of correlations and spectra using triple correlations and bispectra

The relationships between lower and higher order cumulants of deterministic and random continuous signals are presented. This theory is used to develop time- and frequency-domain algorithms for the estimation of correlations (spectra) from triple correlations (bispectra) using acousto-optic processors. Cumulants are of interest because they are insensitive to a wide class of additive noises, including Gaussian noise of unknown covariance. Thus, noise-insensitive correlation (spectrum) estimates can be derived from higher order correlations (polyspectra). The potential for noise insensitivity is examined through the variance of power spectrum estimates based on conventional (second-order) and bispectrum statistics. A proof-of-principle experiment was carried out using an acousto-optic four-product processor to estimate the autocorrelation of a wideband periodic signal. The experimental data are compared with a simulation to validate the results.